September 2026

Changed your mind mid-reply? Type a note while Sensei works. It reads it at its next step, changes course, and tells you what it changed. Your note shows in the reply right where Sensei picked it up.

Product links that point at one option, like a color or a size, now bring back that option's own photo and name, so Sensei matches the exact version you picked.

Paste or drag photos and songs straight into Sensei's message box.

Redo any take from its tile and say what to change, or click the pencil on your last message to edit it and send it again.

See the credits a reply has spent while it works, and get a heads-up in your tab (or a notification, if you turn it on) when Sensei finishes.

The home page has a new look inspired by 80s prime-time TV, starring our own cast. Watch all six episodes of Prime Time season 1, every shot made on Pixel Dojo, and see the four tools behind them.

Our main pages for AI video, AI images, films, music videos, faceless videos, ads, characters, Video Studio, LoRA training and Skills are easier to scan: one idea per section, real examples made with our cast, and the details in a fuller FAQ.

Plans are simpler to compare: each card shows its price, credits and cost per credit, with everything every plan includes in one line underneath.

The new look runs across the whole site: prompting guides, model pages, news, the prompt library, leaderboards and the API platform.

Sensei's plans now show the price of each step, worked out the same way the tool bills it, so the number you approve is the number you pay.

When a job needs more than you approved, Sensei stops and a Continue button picks it up. Nothing past your approval is charged.

Six starter recipes: a 15-second product ad, a UGC testimonial, a product photoshoot, a short film from one line, a music video for your song, and a poster with exact text. Find them on Sensei's start screen and in + → Recipes.

Videos keep rendering through our updates, and Sensei picks them back up instead of stopping.

Make a talking avatar in two steps: describe a voice in Seed Audio, then give OmniHuman a portrait and that clip. A short tutorial walks through it, one click away on the OmniHuman page.

One product, many ads: a two-minute video shows how to keep your saved product and switch between ad styles like UGC, TV spot and unboxing. It opens on your first visit to Marketing Studio and plays on the Marketing Studio page.

Call your saved characters and products by name, draft cheap at 480p, then finish in HD with the same prompt. A 90-second tutorial shows how, one click away in Video Studio.

New films start on Auto, which picks the video engine for you: music videos render on MiniMax H3, the engine that performs your song best, and story films on Seedance 2.5.

Pick an engine yourself, for the whole film or for one shot, and Auto steps aside. Switch back to Auto any time.

If Auto moves a shot to an engine that costs more, you see the new price before anything renders, and nothing is charged until you click again.

The Agent is now Sensei. Describe an ad, a short film or a photoshoot in one message and Sensei makes it: it picks the models our studios use, makes every shot, checks each one and cuts the finished video. Find it in the sidebar, or go to pixeldojo.ai/sensei.

Sensei plans bigger jobs before it starts: a shot list, one shared look for every shot, and a credit estimate. Anything over 40 credits waits for you to press Approve, and the plan stays pinned at the top of the chat, checking off each step as it goes.

It looks at what it makes. Every result gets an honest review against your brief (spelling, faces, hands, and for video six frames across the clip), and a take that misses gets fixed and made again, at most twice.

It can research references on the web, read a model's settings before using it, wait for slow video renders instead of stopping, and cut your clips together with music into one finished video.

When every shot is 1080p, the finished cut is 1080p too.

Save as recipe: when a reply comes out right, save it with blanks for the parts that change, like the product, the logo or the person. Run it again from the + menu and Sensei plans it fresh for the new product, in the same look.

What Sensei makes is a draft for 24 hours. Press Save on anything you want to keep and it goes to your Library, in a folder named after the chat. Finished cuts are saved for you.

Replies keep going if you close the tab or lose your connection, and Recent shows a spinner on any chat that is still working. A Stop button ends a reply at its next step.

Each Sensei reply has a small fee for its thinking, based on what that reply actually used, and it is shown under the reply. Images and videos still cost what the tool costs.

Planning a Faceless Studio episode no longer fails with "The plan came back unreadable" about one time in three. If a plan ever does fail that way, it no longer counts against your daily plans.

Campaign lifestyle shots now show your actual product. Every scene uses your product photo, or the hero image when you don't add one, so the whole set matches.

The campaign launch video now gets made every time, animated from your hero image, instead of being skipped when the hero took a while.

In Video Studio you can tag a saved asset by its first word: @Dewline works for "DEWLINE Rose Water Face Mist".

Sending a Seed Audio clip to OmniHuman now loads reliably, even if the tab opens in the background.

If a page fails to load right after we ship an update, it now refreshes itself once instead of showing an error.

Save buttons, pickers and messages across the site now say Library, the same name as the My Library page where your saved images and videos live. Nothing moved: "Save to Library" keeps a result exactly where "Save to My Media" did.

Faceless Studio now listens to your recorded narration and cuts each scene in the pause before its line, so the picture changes as the narrator moves on instead of a second or two early or late.

Captions are now burned into every Faceless Studio export, timed word by word to the narration.

Make a music video from your own song: a 90-second tutorial shows Film Studio planning, framing and lip-syncing every shot. Watch it on the Music Video Generator page or from the Music videos button in Film Studio.

Voices, sound and music from a prompt: a new tutorial covers narration, full audio scenes and scored tracks. It is one click away on the Seed Audio and Text to Music pages.

New to Faceless Studio? A 90-second video takes a topic to a narrated episode. It opens on your first visit, and it is on the Faceless Studio page too.

The Tutorials section now plays in series order, starting with the welcome tour, so you can watch them one after another.

Film Studio can now color grade your whole film when you export it. Choose from seven looks, including Teal & orange, Golden hour, Moonlight, Bleach bypass and Black & white, and preview each one on a frame from your own film first.

Your clips are never changed by a grade, so you can export again with a different look, or none at all, whenever you like.

New accounts no longer wait up to a day for their first API key to be approved. Your key works as soon as you create it.

The Skills & MCP page is about a third of its old length: what your AI can make, how to connect it, and the full tool list one click away.

The ChatGPT setup steps now match the current ChatGPT: turn on developer mode at chatgpt.com in a browser (the desktop and phone apps don't have that setting), with links straight to the right pages.

Pricing now sits in the top menu, and every studio is linked from the footer.

The home, pricing, studio, LoRA, API and Skills pages now list the current models, real credit costs and what each studio actually does, with answers to common questions on every page.

The example video on the home page now plays when you tap Play.

Qwen LoRA training is now a flat 32 credits at any step count.

In Film Studio, the Export film button no longer runs off the edge of a laptop screen. The export controls now wrap onto a second line when there isn't room.

Faceless Studio's top bar now fits laptop and tablet screens: the price for the next step reads on one line, and long episode titles get the room they need.

Your finished episode no longer stays pinned to the top of the screen while you scroll, so your scenes stay in view on a laptop.

On phones and tablets, vertical episodes show compact scene previews, your episode list shows each episode's full status, and the length picker fits its panel.

The home page opens on a new one-minute film: a giant inflatable dragon loose in Lisbon, every shot made with Seedance 2.5 on PixelDojo. Press play to watch it with sound.

See PixelDojo working inside your AI chat: a short video on the home page and the Skills & MCP page shows one message turning into an image, an ad, a short film and a talking presenter.

A two-minute video shows how to change any image just by describing the edit, upscale it to 4K, and save both steps as a workflow you can run again on any photo. Find it in the Tutorials section.

New to Film Studio? A two-and-a-half-minute video takes one sentence all the way to a finished short: script, storyboard and cast, frames then clips, score and export. It opens on your first visit, you can rewatch it from the button at the top of Film Studio, and it is in the Tutorials section.

A two-minute video shows how to connect PixelDojo to Claude, ChatGPT or Cursor once, then make images, videos and talking presenters just by asking. Watch it on the Skills & MCP page, from the Use PixelDojo inside Claude banner, or in the Tutorials section.

The home page and the Skills & MCP page now show it working as you scroll by: one message in your AI chat becomes an image, a video, an ad and a short film. Click it for the full tutorial with sound.

Just subscribed? If you skip the welcome tour after checkout, it now opens once on your dashboard instead.

New members now get a two-minute welcome tour of PixelDojo right after checkout, covering the tools, the studios and your library. It also leads the Tutorials section on the home page, so you can rewatch it any time.

The community gallery has over 200 new examples across 27 image and video models, from Seedance and Kling to Nano Banana Pro and GPT Image 2, each with the exact prompt that made it. Tap Use this prompt to start from any of them.

The Hailuo Community tab now shows Hailuo videos only. MiniMax H3 videos were mixed in.

New to Character Studio? A two-minute video shows how to create a character once and keep the same face in every scene. It opens on your first visit, you can rewatch it from the Tutorial button in the studio, and it now plays on the Character Studio page and in the home page Tutorials section.

The Tutorials section on the home page and the Tutorials page now play our new walkthrough videos right on the page, with a one-click jump into the tool each one teaches. More are on the way.

New to Canvas? A two-minute video tutorial now walks you from one sentence to a finished cinematic shot. It opens on your first visit, and you can rewatch it any time from the Tutorial button at the top of Canvas.

Pages across the site load much lighter: videos wait until you scroll to them and images come in at the size your screen needs, so the news, guides and AI video pages no longer pull tens of megabytes on open.

Pages appear sooner on your first visit: our headline font now loads straight from our own servers, and background scripts wait until the page is ready to use.

Our home, guide, news and model pages load lighter for new visitors, with Sign in and Sign up ready right away: the sign-in system now loads only where you need it.

Pages in the app no longer jump when a notice or sale banner appears, and the subscribe reminder now sits at the bottom of the screen instead of pushing the page down.

The AI industry news archive can now be browsed page by page, all the way back.

Use case pages show a preview for every example instead of blank tiles, and LoRA model pages show their sample images right away.

Model names across our guides, gallery, prompts, news, contests and landing pages now link straight to that model, so you can go from reading about Seedance 2.5 or Veo 3.1 to making something with it in one click.

Video previews on the API page no longer freeze on a blurry mid-motion frame.

Seedance 1.5 now makes videos in the aspect ratio you pick. Square and vertical requests were coming back widescreen.

Turning audio off in Seedance 1.5 now gives you a silent video.

Hailuo 2.3 videos now finish and show up as soon as they are ready. Some were stuck on "processing" until they timed out, both in the app and through the API.

Similar Images now shows up under every image in the community gallery, including upscales. It had quietly stopped loading.

Gallery and tag searches for everyday words like "portrait" now return matches instead of coming up empty, with the closest matches first.

The home page has a fresh, cleaner look from top to bottom, with a new section on using Pixel Dojo from your own apps and AI assistants.

Pages for our tools, studios, guides, news and the community gallery now share the same calm, easy-to-read design.

News articles are now easy to read in light mode.

Jump back in on your dashboard now shows a clean icon for music and 3D results instead of a broken thumbnail.

Every studio now opens the same clean way, and site notices are calmer and easier to read.

Style pages like Abstract now show community images that actually match the style, and related images in the gallery are closer matches.

Film Studio now plans scenes the way an editor cuts them: an establishing shot, a close-up on whoever is speaking, the reply from the other side, and a reaction where a line lands. Conversations no longer play out in one long static shot.

Before your frames render, Film Studio asks you to cast each character so their face holds from cut to cut. It takes one click, and the description is already written from your script.

Exported films now even out the dialogue, so a line is never loud on one shot and quiet on the next.

Using a character you already have for a role now keeps that role's voice, and their name carries through to every clip.

Marketing Studio now makes ads with Seedance 2.5 by default, for sharper product detail. It runs 6 credits a second at 720p, or 3 for a 480p draft you can finish in 1080p from its card.

Ads keep your product's shape, colors, logo and label text closer to your photos, and every ad style now ends on a clean shot of the product.

Add a reference ad you like, and Marketing Studio copies its hook, pacing and camera moves for your product.

Add captions to any finished ad in one click, for 5 credits.

Importing a product from a store link now shows every photo it finds, so you can keep the three clearest.

Your dashboard now leads with what matters: your latest work first, then this week's sales, then the studios.

Jump back in shows a thumbnail of your latest render from each tool, so you can see where you left off.

Every studio on the dashboard now shows a real example. Hover over one to watch it.

Use Pixel Dojo inside Claude, ChatGPT and Cursor. The home page and dashboard now show you how to connect in a minute.

Ask the Agent for a rough draft of a video and it makes a quick, low-cost 480p preview with Seedance 2.5. Like it? Say so and it renders that exact take in full 1080p.

The same draft-then-upgrade flow works from Claude and other AI assistants connected to Pixel Dojo.

Seedance 2.5 now costs 6 credits a second at 720p and 16 at 1080p (was 7 and 17).

WAN 2.7 image-to-video now costs 2.5 credits a second at 720p and 4 at 1080p (was 3 and 6). WAN 2.2 image-to-video is 0.5 at 480p and 1 at 720p (was 1 and 2).

At 720p, WAN 2.7 Video text-to-video and extend, WAN 2.6 Standard and WAN Reference to Video now cost 2 credits a second (was 2.5, 2.5 and 3).

MiniMax H3 at 2K is 4 credits a second (was 5). H3 Max is 2 at 768p and 1.5 at 480p (was 2.5 and 2).

P Video 2 Pro at 768p in Speed mode is 1 credit a second (was 2), and Grok Imagine Video is 1.5 credits a second (was 2).

Seedream 5 Flash is here: the fastest, lowest-cost Seedream 5, at 0.5 credits an image at any size. Pick Flash in the Model menu on the Seedream 5 page for quick drafts, then switch to Pro for the final image.

Flash is also in Canvas, Consistent Characters, Character Sheets, Character Studio, Film Studio frames and Reality Polisher.

Reality Polisher now shows the right price for Seedream 5 Pro (2.5 credits an iteration).

WAN 2.7 Image and WAN 2.7 Image Edit now cost 0.5 credits an image, and the Pro tier 1.5 (was 1 and 2).

Qwen Image 2 and Qwen Image 2 Edit now cost 0.5 credits an image, and Pro 1.5 (was 1 and 2). Qwen 3 Pro is 1 credit at every size, including 2K and auto (was 2).

Grok Imagine quality images and edits now cost 1.5 credits at 1K and 2 at 2K (was 2 and 3). Face Swap is 0.25 credits (was 1) and Z Image Turbo is 0.25 (was 0.5).

Character Studio, Consistent Characters and Character Sheets charge the same lower prices on WAN 2.7, Qwen Image 2 Pro and Qwen 3 Pro.

WAN 2.7 Video text-to-video and video extend at 1080p now cost 3 credits a second, down from 4.5. A 5 second 1080p clip is 15 credits.

The PixelDojo skills bundle: 11 ready-made workflows for your AI agent, from translating and resizing ads to hook variants, UGC ads, faceless videos and short films. One command installs it in Claude Code, Codex, Cursor and more.

The Skills & MCP page now shows real ads, UGC videos and explainers made through PixelDojo, each with the prompt behind it and a Try in Claude button.

Connect PixelDojo to ChatGPT and Grok Bot, with step by step setup on the Skills & MCP page.

PixelDojo's agent tools now say which ones create media and which only look things up, so your agent knows when to check with you first.

Pick "480p draft" in Seedance 2.5 to preview a shot for the price of 480p. Like the take? The 1080p button on its card renders that exact video at 1080p.

When a video's soundtrack is blocked by the copyright check, you now get a plain explanation of what to change instead of a generic block message. Credits are refunded either way.

Developers can use drafts through the API too: make one with draft set to true, then upgrade it with draft_job_id.

Ask the Agent to upscale with the Creative Upscaler and it's free with your subscription too. The same limits apply: one upscale at a time, up to 20 megapixels each.

The Creative Upscaler is now free with your subscription, on its own page and in Canvas. Upscales run one at a time, up to 20 megapixels each, so the next one starts when the current one finishes.

WAN 3.0 at 720p is now 2 credits a second, down from 2.5, so a 5 second clip is 10 credits.

WAN 3.0 Fast is now 3 credits a second at 720p (was 3.5) and 6 at 1080p (was 7). A 5 second Fast clip at 720p is 15 credits.

WAN 2.7 Image to Video at 720p is now 3 credits a second, down from 4.

Seedream 5 Pro is now 2.5 credits per image at 2K (was 3) and 1.25 at 1K (was 1.5).

Seedream 4.5 and Seedream 5 Lite are back to 1 credit per image, for generating and editing.

Consistent Characters, Character Sheets and Character Studio shots on Seedream cost the same lower prices.

Attach a song in the Agent and ask for a music video. Every shot lip-syncs to its part of your song, and the finished video plays your track.

Ask the Agent for a narrated faceless video on any topic. It writes the script, renders the scenes, adds a voice and music, and saves the finished video to My Media.

Give the Agent a clip to restyle, extend, or put a character into its motion.

The Agent now picks from the same models our studios use, so every task goes to a model proven for it. Films, music videos and episodes from the Agent also work from Claude and other connected AI assistants.

Asking Claude or another AI assistant connected to Pixel Dojo to plan a film from an idea works again. Longer films finish planning in the background, and nothing is charged until you say go.

Cancelling now takes just a few clicks, right here on Pixel Dojo. You see exactly what happens to your plan, credits, library and trained models, and you can undo it any time before your plan ends.

Need a break? Pause billing for 1, 2 or 3 months. We email you a week before your plan restarts on its own.

Switching to a smaller plan now happens right away, and the credits you already have stay on your account.

Credits never expire, even if you cancel. They're ready whenever you come back.

The Subscribe button on tool pages takes you straight to plans again, and after checkout "Back to your tool" returns you to where you were.

The welcome page after subscribing now waits for your credits to land instead of asking you to refresh.

New members get a calmer dashboard: pick what you want to make and you land right in the tool most beginners start with.

Your first creation gets a little celebration, plus a nudge to save it and try animating it in Canvas.

New accounts no longer get months of past What's New updates all at once.

Replies from the Pixel Dojo team on the feedback board now show as the team, not as a community member.

"Report a bug" in the help drawer now opens support chat when you're signed out, instead of failing. Signed in, it tells you exactly what to change if a report is too short or too long.

Each vote on a feedback post counts once, even if you click twice.

Deleting your account now also removes your votes and comments on other people's feedback posts.

The Agent and our AI-assistant connector now make images with Seedream 5 by default, with Grok Imagine as a close second.

Videos now default to Seedance 2.5, and "animate this image" uses your picture as the exact first frame.

If a model turns down a video prompt, the Agent tries once more on WAN 3 and tells you it switched, so fewer requests end in a dead end.

The quick "how likely are you to recommend us" survey now lets you tell us why in your own words, whatever score you give. It no longer pops up while you're subscribing or buying credits.

If you tell us what you love, you can choose whether we may quote you on our site. Quotes are always anonymous, and our privacy policy now explains how they work.

The public feedback board now shows only the posts themselves, never details about who sent them.

Help chat conversations now stay reliably tied to your own signed-in account.

Prepay and save: 6 months of Pro for $135 (500 credits every month) or a full year of Pro Max for $570 (1,100 credits every month), our lowest price per credit.

Upgrading from a prepaid plan adds the credits left on your term right away, so nothing you paid for is lost.

Pro now includes 500 credits a month (up from 400) and Pro Max 1,100 (up from 800), at the same price. Bigger plans get cheaper credits. Current members get the new amount at their next renewal, automatically.

Prepaid 6-month and annual Pro plans now include 500 credits for every month of the term, starting with your next term.

The $250 credit pack now gives 4,400 credits (up from 4,000), and a new $500 pack gives 9,000 credits for subscribers.

Upgrading your plan now replaces your old one, so you are never billed for two plans at once.

Limited-time sales now show up on the homepage and pricing page as picture cards with the discount and end date.

A cleaner look at all five studios on the homepage, with a real example from each.

Model comparison pages now show each model's current price per generation, always up to date, plus any sale running on it.

A new Deals page lists every model on sale right now, with the regular and sale price and the date each deal ends. Sales apply automatically on every plan, no codes needed.

Sale prices now show the before and after credit cost for every model on sale, including video models priced by the second.

The pricing page now shows what a credit is worth on every monthly plan and what popular images and video clips cost, so you can see how far your plan goes before you pick one.

The Canvas model picker now shows the price each model actually charges at its default settings, for every model.

Seedance 2.5 Video Edit in Canvas now holds credits for your clip's real length instead of the 30 second maximum while it runs.

Kling Video 3.0 is now half price: 3 credits a second on Standard and 4 on Pro. Native audio adds 1 a second on Standard and 1.5 on Pro, so a 5 second Standard clip with sound is 20 credits instead of 40.

Kling Reference to Video is 3 credits a second on Standard and 4 on Pro with image references (audio adds 1), and 4 and 5.5 with a reference video. Kling Video Edit is 4 a second on Standard and 5.5 on Pro.

Film Studio, Video Studio, Canvas and AI ad clips that use Kling get the lower price automatically.

WAN 3.0 reference videos now count at the same per second rate as the output, measured before you pay. Auto length jobs now refund the unused seconds once the clip is ready.

Nano Banana Pro is 4.5 credits at 1K and 2K and 7.5 at 4K. Nano Banana 2 is 2.25 at 1K and 4.5 at 4K, and 2K stays 3.

Omni Flash is 3 credits a second at 720p, 4.5 at 1080p and 9 at 4K. Flux 3 Video 1080p is 9 a second, and video to video and extend are 13 at 720p and 16 at 1080p.

MiniMax H3 Fast reference video is 1.75 credits per second of clip, up to 15 seconds, and every H3 model adds 1 credit per reference image past 5. PixVerse V6 1080p with audio is 3.5 a second.

FLUX Video Upscale is now priced by the size and length of the finished video, measured before you pay.

WAN 2.7 Video Edit is now 4.5 credits a second at 720p and 6.5 at 1080p. Happy Horse Video Edit is 5 and 9. WAN 2.6 Flash is 1.5 at 720p and 2.25 at 1080p.

WAN Reference to Video is 3 credits a second at 720p and 4.5 at 1080p (Flash 1.75 and 2.25), and reference videos now count at the same rate for their real length. WAN Character Swap Pro is 5.5 a second, and WAN 2.2 Plus 1080p is 11 per clip.

Veo 3.1 with audio is 12 credits a second on Standard and 4.5 on Fast (8 and 3 with audio off). Kling 2.6 adds 2 a second when audio is on. Kling Avatar Pro is 3.5 a second, Grok Imagine 1.5 is 2.5, PixVerse is 2.25 to 4.5 and Runway Aleph is 10.

OmniHuman is 5 credits a second and charges for the audio you upload, measured before you pay. P Video is now 0.75 a second at 720p on P Video 2 and 0.6 on P Video 1, and jobs with audio are charged for the audio's real length, up to 20 seconds.

Clarity Pro and Portrait Upscaler now price by the size of the finished image, Creative Upscale is 2.5, and Grok Video Extend is 1.5 credits a second of the finished clip. Flux 2 Flex and Max have new size tiers and a small charge per reference image.

Rodin is 12 credits per model, Hunyuan 3D is 15, text to speech is 2 credits per 1,000 characters, and Consistent Characters on Flux Kontext is 1.25 per image. WAN 2.2 Standard 720p is 4.5 per clip.

WAN 2.2 Plus text to video at 480p now renders at 480p instead of 1080p.

Generations that are running while we ship an update now finish instead of stopping partway. If one is ever cut off, your credits come back on their own.

Qwen Image 2 edits now finish in one step and show up right away. They no longer fail as "interrupted".

WAN 3.0 now tells you when you pick a file it can't use. It takes MP4 or MOV video and MP3 or WAV audio.

Fewer Seedance 2.5 video edits fail. When the first try is turned down, we retry it for you as long as your clip fits the length you paid for.

Seedance 2.5 is now 7 credits a second at 720p and 17 at 1080p. 480p stays at 3.

Seedance 2.5 Video Edit and Extend are now 2, 4.5 and 11 credits per billed second at 480p, 720p and 1080p. Extensions now reserve credits for the length the model actually renders, so long extensions no longer come up short.

Seedance 2.0 High is now 5 credits a second at 720p (6 with a reference video) and 11.5 at 1080p.

Consistent Characters and Character Sheets cost 3 credits per image on Seedream 5 Pro and 1.5 on Seedream 4.5. Seedream 5 Lite is now 1.25 credits per image.

Paid with Whop before? Sign in at checkout to use your saved payment methods. We send a quick code to your email to confirm it's you.

If an ad blocker or privacy extension stops checkout from loading, you now get a clear message with a Try again button and a link to finish on Whop, instead of an error page.

MiniMax H3 and H3 Max are now 2.5 credits a second at 768p. H3 at 2K stays at 5, and H3 Max at 480p stays at 2.

MiniMax H3 reference videos now cost 4.5 credits for each second of clip. Upload an MP4 or MOV to PixelDojo and we charge its real length; a pasted link or another format counts as 15 seconds, and the price shows it before you generate.

Kling Avatar and P Video Avatar now charge for the exact length of your audio everywhere you use them, including Canvas, the API and AI assistants. Audio we can't read is turned away before any charge, never billed as a guess.

Kling Video with native audio is now 8 credits a second on Standard and 10 on Pro. Turn audio off to keep the old rates of 6 and 8.

WAN 2.1 Video is now priced by resolution and quality: 2.5 credits a second at 480p and 8 at 720p, or 4 and 14 with the best-quality setting. Clips under 3 seconds bill as 3, and more than 30 steps costs proportionally more.

MiniMax H3 Fast is 1.5 credits a second, plus 10 credits when you attach a reference video. H3 Max Turbo is 1.25 credits a second at 768p and 0.9 at 480p.

Kling Reference to Video with image references is 5 credits a second on Standard, and turning on audio adds 2 credits a second. Pro and video references are unchanged.

Kling multi-shot videos where each shot sets its own length now charge for the full length that is rendered.

AI assistants connected to PixelDojo now see a plain description of every setting on every tool, so they pick the right model and options the first time. For example, they now know a video needs a video model.

Assistants building a film are warned that skipping the music job ID starts, and charges for, a new soundtrack.

The hosted connection for AI assistants now accepts only web links for reference images and clips. Local files still work through the installed connector.

Reality Polisher critiques now actually run on the faster writing model we announced earlier today.

AI assistants can now use Magnific upscaling's full creativity range, from -10 to 10, instead of a narrow slice of it.

Videos you pick from My Media are now measured for their real length. WAN 3.0 and Seedance video edit used to trust the length saved with the clip, which could be off by a second or more, so some requests failed after you clicked Generate.

The My Media tab in the video picker now checks each tool's minimum and maximum clip length, like uploads and links already did.

P-Image Upscale refuses far fewer images, and Film Studio sharpens more of its matched frames instead of falling back to the raw frame.

Film Studio scripts and rewrites come back sharper and faster. Same prompts, better first drafts.

The Agent picks the right tool more often and answers about twice as fast.

Faceless Studio episode plans are more detailed and follow your brief more closely.

Reality Polisher critiques and Film Studio frame checks come back about twice as fast.

Film Studio shot plans come back in less than half the time, with the same care in every shot.

Film Studio scripts and Faceless Studio plans arrive about three times faster.

Prompt enhance, Prompt Studio and the help assistant answer about a third faster.

A shot plan that came back empty now retries on its own instead of asking you to rewrite the script.

Checkout has a fresh look that matches the rest of Pixel Dojo, in light and dark mode.

Buying credits or a plan on your phone is easier. The checkout now uses the full width of the screen.

WAN 3.0 now only offers clip lengths that fit beside your reference video. The 30 second limit covers both, and the picker used to offer lengths that failed after you clicked Generate.

Reference videos over 15 seconds in total are caught before you generate, on the WAN 3.0 page and in Video Studio.

Editing or reusing a fresh Seedream result no longer says the link has expired. Chaining steps in Canvas no longer hits the same false error.

Reference images under 240 pixels are enlarged for you on WAN 3.0, Character Sheets and Consistent Characters, instead of being refused.

A reference file that is no longer available is caught before you are charged, with a clear message to upload it again. Saved drafts drop expired uploads too, and frames you pull from a video now stay usable for a day.

Very large WebP and TIFF images are resized properly before generating, so they no longer fail for being too big.

Film Studio: if sharpening a matched frame fails, that film skips the sharpen for a few hours. Matched frames land faster and cost nothing while it is paused.

Every Character Stylist scenario works again. Most of the choices were being refused.

MiniMax H3 Turbo asks you to remove references before using a start or end frame, instead of failing when you generate.

P Video 2 Pro is live under Generate Videos. Make clips with sound from a prompt or a start image, 5 to 15 seconds long. Add an end frame to land on a target shot. Pick Lowest cost, Speed, or Quality, starting at 1 credit per second.

Video Studio now checks the size of your clip and images before you generate with Seedance. A clip that is too small is refused right away with the exact reason, instead of charging, failing, and refunding.

Switching the model to Seedance after adding a clip runs the same check, so nothing slips through.

Character Studio now shows a Saved shots section for each character. Older shots used to drop out of the Shots view after a busy session, and the Pin as reference button went with them. Every saved shot stays there now, ready to pin.

The prompt box on every tool now grows taller as you type, up to about a third of your screen. Long prompts show around twice as many lines before you need to scroll. You asked for this one.

Tapping into a search box or prompt field on your phone no longer zooms the whole page. Text fields across the site now sit at the size iOS expects.

The tools drawer no longer shows blank placeholder rows when your recent tools cannot load. It now says what to expect instead.

The "new version available" card now appears at the top of the screen instead of covering the Generate button.

The fullscreen image viewer has bigger close, download and save buttons on phones.

My Media page controls fit a phone screen. The page navigation was cut off at the edges.

The Agent chat fits a phone screen. The New chat button was pushed off the edge.

Community Gallery filters scroll sideways on phones so Favorites is reachable again.

Character Studio on phones: larger Shots and Sheet buttons, a bigger close button on every dialog, and form fields that no longer zoom on tap.

Long site notices collapse to two lines on phones with a Show more link, so they take less of your screen.

Setting chips in the composer stay on one line and scroll sideways instead of squeezing and wrapping. Buttons across the studios, gallery, usage report and tutorials are bigger on phones, and the dashboard shows one studio spotlight at a time.

Character Studio: pick a character from a dropdown instead of scrolling a row of names. On phones the header actions fold into one menu and the Shots and Sheet buttons span the row.

Faceless Studio episode list fits a phone screen. Open and Delete had been pushed off the edge. Settings fields no longer zoom on tap.

Film Studio on phones: every button on the storyboard and in the shot editor is easier to tap, and the script and shot fields no longer zoom on tap.

Faceless Studio on phones: the episode header scrolls away instead of covering most of the screen, and scene fields no longer zoom on tap.

Video Studio on phones: the reference and character tiles fit the screen better and the row fades at the edge so you can tell there is more to swipe to.

Canvas on phones: results and start cards no longer hide behind the composer, the model search stops zooming the page, and menus are easier to tap.

Refreshing Canvas now keeps you in the same session instead of quietly starting a new one.

Prompt Studio on phones: the start field is full width instead of a sliver, and the quick-refine chips and version controls are easier to tap.

Agent on phones: a new Chats button opens your past conversations, which had no way in on a small screen. The page also stops scrolling itself to the bottom when you open it.

LoRA trainers on phones: the button to remove an uploaded image now shows on touch screens. It used to appear only when a mouse hovered over it.

Signing up or signing in on an iPhone no longer zooms the page when you tap a field.

Public pages on phones: studio FAQ cards open from anywhere on the card, the features index is a compact two-column list, and small buttons are easier to tap.

Before and after sliders on the upscalers, the gallery and My Media now drag with a finger. They only worked with a mouse before.

My Trainings on phones: bigger action buttons, the thumbnail upload shows on touch screens, and the edit form no longer zooms the page. Pop-up dialogs now scroll on short screens instead of getting cut off.

Holding your phone sideways: the prompt box now sits at the bottom of the screen as a button you tap to open, instead of far down the page. The What's New page shows the latest updates first with older ones one tap away, and pin buttons in the tools menu are easier to hit.

iPad and other touch tablets: buttons that were sized for a mouse are now bigger and easier to tap. Save, remove and frame buttons that only appeared when a mouse hovered over an image now show on touch screens too.

API docs, the contest entry form, gallery comments and the prompt directory are easier to use on phones and tablets: bigger copy buttons and tags, and fields that no longer zoom the page.

Embedding a shared video on your own site now shows a still from the video. Video embeds had been coming up empty.

WAN 3.0 now remembers your reference images, videos and audio when you leave the page and come back.

Clicking a reference on WAN 3.0 drops its tag right where your cursor is, not at the end of the prompt. You can name the same reference more than once.

Enhance on WAN 3.0 now follows the full WAN 3.0 prompt guide even when references are attached, and keeps your reference tags intact.

Seedance Video Edit inserts reference tags at your cursor too, matching Seedance 2.5.

Enhance on Kling, Seedance 2, MiniMax H3, WAN Reference to Video, Grok Video and Happy Horse now keeps your reference tags exactly as you wrote them and never adds tags for files you didn't attach.

Happy Horse and Vidu Q3 get their own Enhance, written from each model's prompting guide instead of the generic one.

MiniMax H3 reference thumbnails are clickable now, matching MiniMax H3 Turbo.

Image editing tools now insert "image 1", "image 2" when you click a reference, the wording every model actually understands, right where your cursor is. Applies to GPT Image 2 and 2.5, Seedream 4, Gemini Image, Flux 2 Flex, Flux Studio, WAN 2.6 and 2.7 Image, Qwen Image, Qwen Image 2, Qwen 3 Pro, Krea, P Image, Luma, Kling Image, Grok Image and Boogu.

Credit prices were adjusted on a handful of tiers so they line up with what each model actually costs to run: MiniMax H3 at 2K is now 5 credits a second, Seedance 2.5 is 3 credits a second at 480p and 6 at 720p, Seedance 2 High is 6 credits a second, Seedance 2 Reference Mini is 2, and WAN Reference to Video at 720p is 3.

Image tiers moved too: Nano Banana Pro at 2K is 4 credits an image, Seedream 5 Pro is 1.5 credits at 1K and 3 at 2K, and Seedream 4.5 Edit is 1.5 credits an image. Every composer, Canvas chip and API doc shows the new price before you generate.

Seedream 4.5 Edit now refunds the full per-image price when part of a batch fails, instead of a flat single credit.

Faceless Studio is live at /faceless-studio. Type a topic, pick a category, a look and a length, and it plans the episode: researched narration, one visual per scene, and a price for every scene before anything renders. Built for faceless channels, educational content and explainer videos.

You own the plan. Edit any line of narration or any visual prompt, switch a scene between a fast 360p draft, a 720p final and a still with motion, and regenerate one scene at a time. Nothing renders until you press Render all.

Record the narration, add a music bed, and export a captioned episode with the voice on top and the music under it. Every clip can be previewed before you export.

Planning is included with your subscription. Scenes cost what the engine costs, 1 credit a second at 360p and 2 at 720p, and the header always shows what you have spent so far.

Six looks with a still for each, three real example episodes on the home page, and a list of your own episodes to come back to.

Video Studio is a single page for four video jobs. Pick Make, Change, Move or Extend, drop in what you have, and press Generate. We pick a good model for each job, and you can switch to another that can do it.

Real examples sit right under the headline. Drag the before and after slider on a Change or Move example to see exactly what it did, then press Try this to start from it.

Your saved characters, products and outfits attach as tiles. Click one to write it into the prompt as @Name, and pick which of its photos get sent.

Every finished clip can go straight back in. Extend it, or change something inside it, without leaving the page.

The price on the button is the price of the clip you set up, including the length of any video you attach.

FLUX Video Edit is live at /flux-video-edit under Change a video. Upload a clip up to 15 seconds, describe the change in plain language, and get it back with the same length, framing, and audio.

2 credits per second of your source clip, 720p output. No settings to tune: pick a clip, write the edit, press Edit.

Available in Canvas as a video edit step and on the public API as the flux-video-edit model ID.

The prompt box no longer jumps up and down while you type. The safety check used to flash a status line under the prompt on every pause, which pushed your text around and lost your place. It now runs quietly and only shows a message if something is actually blocked.

GPT Image 2.5 is live at /gpt-image-2-5 in two flavors. Sunburst is the premium model: it keeps a reference subject faithful, makes localized changes without disturbing the rest of the frame, and handles dense layouts. Flare returns the same quality tier at about half the wait.

Generate and edit live on one page. Add up to 8 reference images and the same prompt box becomes an edit instruction, with an optional mask for inpainting.

0.5 credits at Low, 1 at Medium, 2 at High, with 4K doubling Medium and High. Edits with references add 0.5 credits per image. Both variants cost the same, so pick Flare when you want speed and Sunburst when fidelity matters.

Available in Canvas as a generate step with model, size, quality, and format controls, and on the public API as the gpt-image-2-5 model ID.

A prompting guide with real generated examples is at /guides/gpt-image-2-5-prompting-guide.

MiniMax H3 Max Turbo is the distilled, faster H3 Max. Text or a start frame to 5 to 15 second clips with synced audio at 768p for 1 credit per second, or 480p for 0.6. A 5 second 480p clip is 3 credits.

MiniMax H3 Fast is the quickest H3 and the cheap way to preview a reference shot: it takes the same image, video and audio references as H3, renders 480p, and bills 1.1 credits per second. Both are in the MiniMax H3 tool, in Canvas, and in the API as minimax-h3-max-turbo and minimax-h3-fast.

Saving an image or video on iPhone and iPad works again. If the share sheet does not open, the file now downloads straight to your device instead of doing nothing.

Every download button now works the same way, including My Library, the batch Download bar, the community gallery, and Film Studio exports.

Saving from inside an app browser like Instagram or TikTok now tells you to open the page in Safari or Chrome if the file did not land.

Gemini Omni Flash is now version 1.1. Pick your resolution from a cheap 360p draft up to 1080p and 4K, all with the same built-in sound. 360p is 1 credit a second, 720p stays at 2, 1080p is 3, and 4K is 6.

Add an end frame next to your start image and the model animates the journey between the two stills. Great for reveals, orbits, and clean loops.

Reference clips join reference images. Hand it up to three short videos alongside up to seven stills to carry a character, a style, or a motion into the new shot.

Extend any clip. Hit Extend on a result, or bring your own video up to 30 seconds, and add 3 to 10 seconds of what happens next. The whole longer clip comes back in one piece.

Editing a video now keeps its original length, and you choose the output resolution. The old Continue button is replaced by Extend, which works on any clip, not just ones made here.

P Video Edit is live under Edit Videos. Describe the change you want on a clip up to 15 seconds, add up to 4 reference images, and keep the original audio. Draft mode gives a quick preview at half the credits.

WAN 2.2 Image to Video now takes an optional end frame. Add a second image on the End tile and the clip moves from your start image to it.

On WAN 3.0, WAN 2.2 and 2.7 Image to Video, Face Swap, Qwen Image Edit, Z-Image and MiniMax H3 Turbo, the words "girl" and "boy" in your prompt now reach the model as "woman" and "man". Your prompt is still checked exactly as you wrote it first; this only changes what the model is asked to render.

WAN 2.7 Image to Video runs on a new backbone with fewer content restrictions. Same start-image flow, 720p or 1080p, clips from 5 to 15 seconds.

WAN 2.2 has a new Image to Video mode, now the default when you open the page. Turn a photo into a 480p or 720p clip from 1 credit a second.

The previous WAN 2.2 image mode is now called Image to Video LoRA, so it is clear which mode takes your trained LoRAs, end frames and the Plus and Extended tiers.

WAN 3.0 now renders in 2K and 4K. Pick the resolution from the chip next to the prompt. Both work in Standard and Fast mode, with native audio and clips up to 30 seconds.

WAN 3.0 now follows your prompt literally by default. Prompt rewriting is off unless you turn it on in More Options, so what you type is what gets rendered.

1080p Standard is now 5 credits per second. Every other WAN 3.0 price is unchanged. You always see the exact cost on the Generate button before you run.

Seedream 5, Seedream 5 Lite, Seedance 2.5 and Seedance 2.5 Video Edit now send your prompt to the model word for word. No automatic rewriting behind the scenes, so what you type is what gets rendered.

Seedream 5 Lite blocked far fewer stylized character prompts in our tests once the rewriting step was removed. If a prompt used to fail on Lite but work on Seedream 5, try it again.

Uploading a reference image now closes the picker as soon as the image is added. Before, the window stayed open with no clear sign the upload had landed.

MiniMax H3 Turbo is here. It makes video with native synced audio faster, and for fewer credits than MiniMax H3.

Choose from three speed and quality tiers at 480p or 720p, for clips of 5 to 15 seconds. The fastest tier costs the least, so you can iterate cheaply and save your credits for the take you want.

Bring up to 9 reference images, a start and end frame, or reference video and audio to guide the shot.

If your prompt mentions @Image1, @Video1 or @Audio1 but you have not attached that reference, Seedance 2.5 now tells you before it takes any credits. It used to accept the run, invent the missing subject, and fail minutes later.

Video Upscaler is available again through the API and MCP agents. Send a clip and get it back at up to 4K. It does not have a page in the app for now, so look for it in the API docs.

Images and videos generated through the API or an MCP agent are now available for 1 hour after they finish, down from 24 hours. Every job response tells you the exact expiry time. Download anything you want to keep.

The API and MCP save action has been removed. Files can no longer be copied into My Media with an API key. Saving to My Media stays a signed-in action inside Pixel Dojo. Uploads for reference images still last 24 hours.

A blocked generation no longer counts toward an automatic account suspension. Blocked attempts are still recorded and reviewed, and any action on an account now comes from that review rather than a strike counter. Confirmed violations involving minors still close the account. If your account was suspended under the old rule, that suspension is being lifted today.

The Terms of Service, Privacy Policy, Trust & Safety page, and our anti-trafficking, content safety, and content removal policies now describe this review process.

Version 3 of our Terms of Service is now in effect. It adds rules for developers who build on our API and MCP server, notes that some models carry their own usage rules, and asks you to report misuse of an integration. The next time you open Pixel Dojo we will ask you to accept it. Nothing about your account, your credits, or your saved work changes.

Visitors from the EU, UK, and Switzerland now see a cookie banner before any analytics or ad measurement loads. Everyone can change their choice at any time from Cookie settings in the footer or in Settings.

If your browser sends a Global Privacy Control signal, we treat it as an opt-out of ad measurement automatically.

When a purchase is refunded, we now remove exactly the credits that purchase added. Never more, and never below zero.

We rewrote our Terms of Service, Privacy Policy, and Refunds & Billing policy in plain language, and brought every other policy page in line with them, from the Cookie Notice to the Sub-Processors list. They now name the company you are dealing with, spell out one clear refund rule, confirm that your credits roll over for as long as you are subscribed, and describe exactly how reports and appeals are handled.

The next time you open Pixel Dojo you will be asked to read and accept the updated Terms once. It takes one click, and nothing about your account, your credits, or your saved work changes.

Face Swap has returned to Characters & Faces. Pick a target photo and the face you want in it, choose face-only or full head, and the swap keeps the original pose, lighting, and background. One credit per swap.

Both photos in a face swap now pass the same safety check as every other uploaded image. Before your first swap, the tool asks you to confirm once that everyone in your photos is an adult who agreed to be there.

Sending a photo to a tool with Use in from My Media no longer refills the other image slot from an old draft. Only the photo you picked lands in the composer.

The Creator Agreement replaces the old content toggle in Settings. It links to the Terms of Service and says plainly what you confirm: you are 18 or older, you hold the rights to everything you upload and have consent from anyone shown, and automated abuse checks can lead to suspension or termination. Everyone accepts it once, even if the old toggle was already on.

Happy Horse video clips start at 3 seconds now. The 2 second option always failed after charging and refunding your credits, so it is gone.

WAN 3.0 now checks your reference image size up front. Images smaller than 240x240 pixels get a clear message naming your image's actual size, before any credits move.

WAN 2.7 and WAN 2.6 Image Edit now accept small reference images. Anything under 240 pixels on a side is enlarged for you before the edit runs, so a tiny character portrait no longer fails every time and hands your credit back.

Z-Image Turbo shows a plain message when a generation takes too long, instead of a raw error code.

The MiniMax H3 prompting guide now covers H3 Max: the same prompt run on both models side by side, what a start frame does to the bill on each, and a plain rule for when to pick which.