Which AI model should I use? Pick the right one first.

Best at

• Crowded prompts where every object, color and position has to land
• Photoreal close-ups with true skin, fabric and glass detail
• Posters and storefronts with several lines of exact text in one image
• Editing your own photos, with up to 10 reference images

Pick another model when

• You need a 4K image or custom sizes up to 4096 pixels: Seedream 4.5
• You are designing a logo or packaging where every word must be perfect: Ideogram 4.5
• You want to mix more than 10 reference images: Nano Banana Pro and Nano Banana 2

Tips and quirks

• Auto sizes come back square from text. Pick a landscape or portrait size instead.
• Put each line of text in double quotes and end with "the text must be exact".
• Concrete physical detail beats hype words. "Water streaming off the brass helmet" works, "amazing" does not.
• The engine picker also holds Flash and Lite. Find the idea there, then rerun the prompt on Pro.

Best at

• Polished, good-looking photos from short, simple prompts
• Fashion, lifestyle and portrait shots with natural skin texture
• Sharp 4K output for print and wide atmospheric scenes
• Editing photos with up to 14 references while keeping the polished look

Pick another model when

• Your prompt lists many objects that all have to appear exactly as described: Seedream 5
• You need a headline or label spelled perfectly: Ideogram 4.5

Tips and quirks

• Trust its built-in taste. You rarely need to name a color palette.
• Use Auto 4K for print, fog, water and foliage. Auto 2K is plenty for social posts.
• Need an odd shape? Set a custom width and height up to 4096 pixels.
• Add up to 14 reference images and say which one carries the subject and which the style.

Best at

• Trying many versions of an idea quickly before you pick one
• Copying an edit from one example pair onto a new image
• Large 3K images that blend up to 14 reference photos

Pick another model when

• The final image needs the sharpest detail and the cleanest text: Seedream 5
• You need a full 4K image: Seedream 4.5

Tips and quirks

• Lite is an engine inside the Seedream 5 page. Choose it from the model picker there.
• Attach a before and after, then ask for the same change on image 3.
• Write full sentences, not tag lists. "A woman walking under a parasol" beats "woman, parasol".
• Say what must stay the same when you edit, like the framing and the lighting.

Best at

• Edits that follow your instruction to the letter, like "remove the hat"
• Generating and editing on one page, with up to 3 reference images
• Signs, chalkboards and labels spelled right when the words are in quotes

Pick another model when

• You want to combine more than 3 reference images: Qwen Image 2.1
• You want to use a LoRA you trained: Qwen Image 2.1
• A poster or logo needs dense, exact typography: Ideogram 4.5

Tips and quirks

• Write edits as commands ("make his shirt red"), not as a caption of the finished picture.
• Call references image 1, image 2, image 3 in the order you attached them.
• On Auto size the result takes the shape of the last image, so attach that one last.
• Leave Prompt Enhancement off for exact edits. Turn it on only for a three word idea.

Best at

• Crisp type and fine small print on posters, rendered natively at 2K
• Editing and combining up to 10 reference images in one prompt
• Running up to 3 LoRAs you trained, like your face plus a style

Pick another model when

• You want quick drafts or lots of images: Qwen Image 2.1 Turbo
• You need a 4K image or a real place looked up: FLUX 3 Image
• Every word on the design has to be exact: Ideogram 4.5

Tips and quirks

• This page holds two versions. Pick the full model for final art and the finest small print.
• The full model reads a negative prompt. Turbo ignores it.
• Quote every word and say where it sits, like "across the top". Name a type style too.
• Train your LoRA with a made up trigger word and write it where the subject goes.

Best at

• Fast native 2K images for drafts, product shots, food and interiors
• Free to run on any plan with Relax
• Edits with up to 10 reference images and up to 3 of your own LoRAs

Pick another model when

• You want the most detail and the finest small print: Qwen Image 2.1
• You write short prompts and want a full, rich scene back: FLUX 3 Image Klein
• You want to keep a lot out with a negative prompt: Qwen Image 2.1

Tips and quirks

• Turbo is the default on the Qwen Image 2.1 page. Switch the model to get the full version.
• Shape the idea at 1K, then render your favorite at 2K or on the full model.
• Turbo skips the negative prompt, so describe what should be there instead.
• Give each reference a job in the prompt: the product, the setting, the face.

Best at

• Text-heavy layouts like infographics, slides, magazine covers and comic panels
• Long, detailed prompts with many text blocks, followed in order
• Two tiers on one page, Standard and Pro, up to 6 images per run

Pick another model when

• You want to combine more than 3 reference images or use your LoRAs: Qwen Image 2.1
• An edit has to follow your instruction exactly: Qwen 3 Pro
• A logo or label has to be letter-perfect: Ideogram 4.5

Tips and quirks

• List text blocks top to bottom and name a font style for each one.
• Number the steps in an infographic prompt. The model keeps the sequence.
• For edits, say what stays the same first, then what changes.
• Prompt Enhancement is off by default so your exact wording is kept.

Best at

• Posters, slides and icon sets that come back exactly as you laid them out
• Keeping the same person or product across new scenes from one photo
• Precise edits that change only what you list

Pick another model when

• You want the model to add its own flourishes and props: Nano Banana Pro and Nano Banana 2
• You are making a logo or packaging where the wording must be flawless: Ideogram 4.5
• You want a detailed photoreal hero shot from a long, crowded prompt: Seedream 5

Tips and quirks

• Write it like a design brief. Say "a finished poster" first, then the structure, copy and finish.
• Use numbers. Four panels, a three by three grid and two colors are followed closely.
• For edits, write "exactly three changes", list them, then list what must stay.
• Call a photo "the identity reference" and name what to keep, like the face, braid and freckles.

Best at

• Dense infographics, diagrams and magazine covers with Nano Banana Pro
• Blending up to 14 photos into one new scene
• Everyday images and edits at 1K, 2K or 4K in ten aspect ratios

Pick another model when

• You want a logo or label where every letter is perfect: Ideogram 4.5
• You want to paint a mask and change only that area: GPT Image 2.5

Tips and quirks

• The picker holds the whole Nano Banana family. Nano Banana 2.1 is the default.
• Switch to Nano Banana Pro for heavy text, step-by-step worksheets or code on a screen.
• Nano Banana 2 adds its own props and decoration. Use it when you want a playful, busy look.
• Give every reference image a job, like "the fabric sample is the texture", then describe the new scene.

Best at

• Full, detailed scenes from a one line prompt
• Free to run on any plan with Relax
• Edits and composites with up to 10 reference images, up to 4 images per run

Pick another model when

• You need 4K or a real product or place looked up: FLUX 3 Image
• You want to use a LoRA you trained: FLUX.2
• Every word on the design has to be exact: Ideogram 4.5

Tips and quirks

• Keep prompts short. We expand them into a full scene before it renders.
• Anything you do name, like a color, lens or material, stays in the picture.
• Turn Prompt Upsampling off only when you paste a complete scene description.
• On Auto the result keeps the shape of your first reference image.

Best at

• Photoreal scenes and product shots from a quick draft up to 4K
• Real landmarks and products, looked up on the web before it draws
• Composites from up to 10 reference images, like a product placed in a room

Pick another model when

• You want lots of everyday images or drafts: FLUX 3 Image Klein
• You want to use a LoRA you trained: FLUX.2
• Every word on a poster or label must be exact: Ideogram 4.5

Tips and quirks

• There is no negative prompt or seed. Write "an empty beach", not "no people".
• Open with the image type and medium, then subject, setting, light and framing.
• Put a hex code right next to the one object it colors.
• Turn Web Grounding off for invented scenes. It only helps with real things.

Best at

• Photoreal portraits and product photos with natural hands and anatomy
• Brand colors placed exactly, using hex codes tied to each object
• Eight FLUX.2 versions on one page, including two that run your own LoRAs

Pick another model when

• You need a 4K image or a real place looked up: FLUX 3 Image
• You want fast everyday images without picking a version: FLUX 3 Image Klein
• You want to transform one photo with a strength slider: FLUX Image to Image

Tips and quirks

• Flex takes the most references (10). Pro and Max take 8, Dev 4.
• Put the must-haves early in the prompt. FLUX.2 weighs the first words most.
• Skip negative prompts. Describe what you want to see instead.
• Pick Dev + LoRA or Klein 9B + LoRA to load a LoRA. Other versions ignore it.

Best at

• The classic FLUX.1 family, from fast Schnell to detailed 1.1 Pro Ultra
• Character and style LoRAs, including stacking several on Dev Multi LoRA
• Kontext edits that change one thing and keep the rest of the photo

Pick another model when

• You want sharper detail and closer prompt following: FLUX.2
• You want to combine many reference images: FLUX 3 Image
• You want to restyle one photo with a strength slider: FLUX Image to Image

Tips and quirks

• Krea Dev is the everyday pick. It avoids the glossy, oversaturated look.
• Use Kontext Pro or Max for words in the image or to edit a photo.
• Only Krea Dev and Dev Multi LoRA load LoRAs. Pick one of those for your trained models.
• Kontext combines at most 2 images in one edit.

Best at

• Restyling a photo while keeping its layout, like turning it into an oil painting
• Applying your trained FLUX LoRAs to a photo you upload
• Fine control over how much changes, with one strength slider

Pick another model when

• You want one precise change, like removing a hat: Qwen 3 Pro
• You want to combine several images into one: FLUX 3 Image Klein
• You want to make an image from text only: FLUX.1

Tips and quirks

• Lower strength keeps more of the original. Higher strength follows the prompt more.
• Start near the default of 0.8, then nudge down if the subject drifts.
• Describe the whole finished image, not just the change. It redraws everything.
• Put your LoRA's trigger word in the prompt so the LoRA takes effect.

Best at

• Edits that change one thing and leave the rest of the photo untouched
• Storyboards and multi-panel layouts where the same character holds
• Natural light on skin, clay and wet surfaces, up to 4K

Pick another model when

• You need a logo, label or sign with perfect lettering: Ideogram 4.5
• You want to combine more than 8 reference images: Nano Banana Pro and Nano Banana 2

Tips and quirks

• Use Flare to try ideas quickly, then rerun the winner on Sunburst.
• End every edit with what stays: "keep the table, lighting and camera exactly the same".
• Ask for space you will need later, like "negative space on the right for copy".
• For masked edits, white marks the area to change and black is kept.

Best at

• Quick moodboards with many variations of one idea
• Atmospheric landscapes, documentary portraits and small product shots
• Exact pixel sizes for banners and posts, plus edits with up to 5 images

Pick another model when

• Words on a sign or package must be exact: Ideogram 4.5
• You want polished hero shots with fine detail: Seedream 5
• You want photoreal images up to 4K: FLUX 3 Image

Tips and quirks

• Each run makes one image. Run it a few times to compare directions.
• Pick Custom size to set exact width and height, like 1200 by 630.
• Name the medium, like "storybook watercolor", or it leans photoreal.
• Atmosphere words such as "twilight" or "aurora overhead" land strongly.

Best at

• Up to 10 quick variations of an idea in one run
• Moody, cinematic and stylized looks from short prompts
• Fast edits to one photo: restyle, relight or swap the background

Pick another model when

• You want to blend several reference photos into one image: Nano Banana Pro and Nano Banana 2
• The image has to carry exact words: Ideogram 4.5

Tips and quirks

• Switch to the Quality or Imagine 2 tier for sharper detail and better text.
• Name a mood you can picture, like "Blade Runner mood", instead of "dark".
• One clear subject on a defined background beats a busy scene.
• Results vary run to run. Try the same prompt again before you rewrite it.

Best at

• Posters, logos, packaging and signs with clean, correctly spelled text
• Changing one word or price on a finished design with Precise Edit
• Brand sheets with an emblem, wordmark, tagline and named color swatches

Pick another model when

• You want a photoreal portrait or scene with no text: Seedream 5
• You need the same person across many images from their photos: Nano Banana 2.1
• You want to combine more than 5 images: Nano Banana Pro and Nano Banana 2

Tips and quirks

• Put the words in quotes near the start. Give each line its own place and style.
• End with "the only words on the cover are..." so it adds no extra lines.
• Describe colors and fonts in plain words. Hex codes and font names can show up in the image.
• Say what you want, not what you don't. "An empty room" works better than "no people".

Best at

• Natural photoreal portraits and lifestyle shots
• Running up to 3 LoRAs, including ones from our Z-Image trainer
• Short signs and labels in English or Chinese, written in quotes

Pick another model when

• You want to edit or combine photos you already have: FLUX 3 Image Klein
• You need an ultra wide 21:9 frame: FLUX 3 Image
• Your design has lines of text that must be exact: Ideogram 4.5

Tips and quirks

• There is no negative prompt. Say what you want instead of what to avoid.
• Write full sentences, not tag lists. It reads natural language best.
• Name exact materials, like "brushed brass", rather than "premium materials".
• Frames are limited to 1:1, 16:9, 9:16, 4:3 and 3:4.

Best at

• Polished, art-directed images with very little prompting
• Concept art, fashion editorials and anime when you name the style
• Your own look or character through a Krea LoRA trained on PixelDojo

Pick another model when

• You want to edit an existing photo: Nano Banana 2.1
• You need a paragraph of text or dense type: Ideogram 4.5

Tips and quirks

• Creativity sets how literal it is. Raw follows your words, High takes more liberties.
• Style references change palette, texture and mood, not what is in the picture.
• Quality words and named photographers or magazines help here, like "Vogue magazine style".
• Explore on Turbo, then lock the seed and rerun on Large.

Best at

• Editorial portraits with fine skin and fabric detail
• Ultra-wide 21:9 landscapes and small figures in vast scenery
• Batches of up to 9 images from one prompt to pick from

Pick another model when

• You need a headline or more than a few words of text: Ideogram 4.5
• You want to work from several reference images at once: Nano Banana Pro and Nano Banana 2
• You need a 4K image: Seedream 4.5

Tips and quirks

• It comes with a short negative prompt already filled in. Keep yours short too.
• Use camera words for photos and name the medium for art, like watercolor or gouache.
• Use 1K to explore, then 2K for the final where fog, water or fabric detail matters.
• Run a batch of 4 to 9 early, then refine the best one.

Best at

• 5 to 15 second clips with synced sound from a prompt, a photo, or references
• Keeping up to 9 people, products or places consistent in one shot
• Opening on your photo and landing on an end photo you choose

Pick another model when

• You want sharper 2K output instead of 720p: MiniMax H3
• You need one take longer than 15 seconds: WAN 3.0
• You want a song or line played exactly as recorded, with lip sync: P-Video 2 Pro Reference

Tips and quirks

• Tag references in the prompt as <Picture 1>, <Picture 2>, in upload order.
• With a start photo, spell out the camera move and how the shot ends.
• Hear mumbled speech? Add "Nobody speaks or sings, only music and ambient sound."
• Subscribers can switch to Relax mode for unlimited 5 second clips at 720p.

Best at

• Readable on-screen text, like title cards spelled exactly as typed
• Sharp 2K clips up to 15 seconds with stereo sound in the same file
• Mixing up to 9 images, 3 videos and 3 audio clips as references

Pick another model when

• You want lots of quick tries at 720p before the final: MiniMax H3 Turbo
• You need references and an exact start frame in the same clip: P-Video 2 Pro Reference
• You need one take longer than 15 seconds: WAN 3.0

Tips and quirks

• Type the exact words in quotes for on-screen text, then add "no other text, no subtitles".
• Name one camera move. Left open, it often cuts between two setups on its own.
• Start and end frames can't be combined with references here. Pick one approach per clip.
• The model menu also has H3 Max and Max Turbo (768p, frames only) and H3 Fast (480p, references).

Best at

• Single takes up to 30 seconds with native dialogue, music and sound
• Up to 10 reference images, plus reference video for motion and audio for voice
• Up to 4K, or let it pick the frame shape and length for you

Pick another model when

• You need exact words on screen, like a title card: MiniMax H3
• You want to continue a clip you already have: FLUX 3 Video
• You want to restyle existing footage with an instruction: WAN 2.7 Video Edit

Tips and quirks

• Split long clips into stages, each with one change and an end state you could point at.
• Say "One continuous take" if you don't want it to cut between camera setups.
• Put spoken lines in quotes, or write "No dialogue throughout the entire video" to avoid made-up speech.
• Use either references or start and end frames, not both. Give each @Image 1, @Image 2 one job.
• For two or more people, start from one still of the whole cast. Separate photos can mix up faces.

Best at

• Extending a 2 to 10 second clip into a longer continuous shot
• Animating a photo in time with your own voice or music track
• Text or image to video with sound at 720p or 1080p, 2 to 15 seconds

Pick another model when

• You want one take up to 30 seconds from the newest WAN: WAN 3.0
• You want to change how an existing clip looks, not continue it: WAN 2.7 Video Edit
• You want to extend a clip in 1080p or 21:9 ultrawide: FLUX 3 Video

Tips and quirks

• For image to video, prompt the motion and camera. The photo already sets the look.
• In extend mode, the new length must be longer than your source clip.
• Leave the default negative prompt as is. It already blocks subtitles and watermarks.
• Use 1080p for fog, water and fine texture. 720p is plenty for action and faces.

Best at

• Dropping the same characters into a new scene from 1 to 5 photos or clips
• Two or more characters acting together in one shot
• Multi-shot clips that keep faces steady across the cuts

Pick another model when

• You need more than 5 references or a clip longer than 10 seconds: WAN 3.0
• You want up to 9 references with reference audio too: MiniMax H3 Turbo

Tips and quirks

• Call subjects character1, character2 in the prompt, in the order you uploaded them.
• WAN 2.7 in the model menu holds characters best. WAN 2.6 Flash is the quick option for tests.
• Switch the shot type to multi-shot when you want a short sequence with cuts.
• Use clean, well lit photos with plain backgrounds. Busy backdrops leak into the scene.

Best at

• Restyling a 2 to 10 second clip with a plain instruction
• Bringing in a look, outfit or object from up to 4 reference images
• Keeping the original soundtrack while the picture changes

Pick another model when

• You want to continue a clip rather than change it: WAN 2.7 Video
• You want a brand new clip from a prompt: WAN 3.0

Tips and quirks

• Set audio to Keep Original when the clip has dialogue or music you need.
• Say what must stay the same, like the motion, framing and the main person.
• For a restyle, name the medium and palette, such as watercolor in warm coral tones.
• Trim sources to 10 seconds or less before uploading.

Best at

• One continuous scene up to 30 seconds, with sound, from a single prompt
• Mixing up to 30 photos, 10 clips and 10 audio tracks as references in one shot
• Dialogue, ambience and effects in 11 languages, made together with the picture

Pick another model when

• Your cast must look exactly like their photos, or a label must be spelled right: Kling 4.0
• You want to change or extend a clip you already have: Seedance 2.5 Video Edit
• You want a 4K finish on a clip up to 15 seconds: Seedance 2

Tips and quirks

• Tag each reference as @Image1, @Video1 or @Audio1 and say what each one adds.
• Add a short exclusion per reference, like "do not use the image background", so nothing extra leaks in.
• Use a start frame or reference photos, not both. A start frame also sets the clip's shape.
• With reference photos, set the aspect ratio yourself. Vertical needs 9:16 picked, or it frames wide.
• Want only natural sound? Add "no background music" or it may score the scene.

Best at

• Swapping a product, person or background inside a clip you already have
• Adding new footage before or after a clip, for a result up to 30 seconds
• Changing the sound, like removing the music, while the picture stays the same

Pick another model when

• You want a brand new clip from a prompt or photos: Seedance 2.5
• You want a person from a photo to perform another clip's moves: Kling Motion Control

Tips and quirks

• Pick Edit or Extend first. Edit keeps your clip's length and shape.
• Skip words like "continue" in an Edit prompt or "replace" in an Extend prompt. The mode handles that.
• Your source clip is always @Video1 and must be 4 to 30 seconds long.
• Name the one change, then list what must stay, like the face, camera move and dialogue.

Best at

• Putting your own character into a new scene from up to 9 photos
• Copying a dance or camera move from up to 3 reference clips
• Timing the action to a song or voice track you upload

Pick another model when

• You need more than 15 seconds or more than 9 reference photos: Seedance 2.5
• You want the clip to open on an exact first frame: Seedance 2
• Faces must match the reference photos as closely as possible: Kling 4.0

Tips and quirks

• Needs at least one photo or clip. An audio track alone is not enough.
• Tag inputs as [Image1], [Video1] or [Audio1] and say what each one drives.
• 1080p and 4K need the High quality tier. Fast and Mini stop at 720p.
• Reference clips can add up to 15 seconds in total, so trim long videos first.

Best at

• Punchy action with real weight: chases, fights, falling debris
• Crisp 4K output on the High quality tier
• Animating between a start frame and an end frame you choose

Pick another model when

• You want a length other than 5, 10 or 15 seconds, up to 30: Seedance 2.5
• Your character or product comes from reference photos: Seedance 2 Reference

Tips and quirks

• Clips come in 5, 10 or 15 seconds only.
• Mini and Fast stop at 720p. Switch to High for 1080p or 4K.
• Use time brackets like [0-5s] to give each shot its own camera and sound.
• Pack in textures, light and sounds. Sparse prompts get generic motion.

Best at

• Natural, atmospheric motion like wind in hair, moving water and shifting light
• Quick clips from 1 to 15 seconds in seven aspect ratios
• Animating a still with matching sound on Grok Imagine 1.5

Pick another model when

• You want sound on a clip made from text alone: Veo 3.1
• You need 1080p or a clip longer than 15 seconds: Seedance 2.5

Tips and quirks

• The standard model is silent. Pick Grok Imagine 1.5 with a start image for sound.
• Name one camera move. Asking for two at once blurs the motion.
• For image-to-video, keep it short and describe only what moves.
• On 1.5, describe the motion first, then the sounds you want on an AUDIO: line.

Best at

• Quick 5 to 15 second clips with sound from a prompt or a start image
• Clips that land exactly on an end frame you choose
• Seven aspect ratios, including 3:2 and 2:3

Pick another model when

• You want sharper 2K or 1080p output: MiniMax H3
• You need one take longer than 15 seconds: WAN 3.0
• You want to keep a person or product consistent from photos: P-Video 2 Pro Reference

Tips and quirks

• Pick Quality mode for the best motion and detail. Speed is the default.
• The prompt upsampler rewrites your prompt. Set it to Off to keep your exact wording.
• Attach any reference image, video or audio and the page switches to Reference mode.
• Output tops out at 768p. Use 480p for fast tests of an idea.

Best at

• Lip sync to your own song or line, kept exactly as recorded
• References and start or end frames together in the same clip
• Up to 9 images, 3 video clips and 3 audio clips as references

Pick another model when

• You want sharper 2K output from references: MiniMax H3
• You need one take longer than 15 seconds: WAN 3.0
• You have no references and just want a quick clip: P-Video 2 Pro

Tips and quirks

• This is the Reference mode of P-Video 2 Pro. It turns on when you attach a reference.
• With one audio file attached, tick "Use as soundtrack" to keep it exactly and sync the lips.
• Refer to references by what they show, like "the woman in the red coat".
• Start and end frames count toward the 9 image limit. Quality mode is not offered here.

Best at

• Believable motion with real weight, natural arcs and folding fabric
• Two or three numbered shots cut inside one 10 to 15 second clip
• Placing up to 9 reference photos in a scene in Reference to Video mode

Pick another model when

• You need one take longer than 15 seconds: WAN 3.0
• You want to extend a clip you already have: WAN 2.7 Video
• You need exact words on screen: MiniMax H3

Tips and quirks

• Number your shots (Shot 1, Shot 2) and give each its own camera angle and action.
• Write plain prose. Skip tag lists, weights and JSON.
• Crop your start image to the shape you want. The clip follows the image, not a setting.
• In Reference mode, call subjects character1, character2 in upload order.

Best at

• Polished, cinematic shots up to 8 seconds at 1080p with sound
• Spoken lines with lip sync, written right in the prompt
• Smooth transformations between a first frame and a last frame

Pick another model when

• You need a clip longer than 8 seconds: Kling 4.0
• One character must look the same across many clips: Kling 4.0
• You want a square, 4:3 or 21:9 frame: Seedance 2.5

Tips and quirks

• Only 16:9 and 9:16 are offered. Clips are 4, 6 or 8 seconds.
• The page opens on Fast. Reference photos need Standard, set to 16:9 and 8 seconds.
• Write dialogue as speaker, colon, then the line in quotes. Keep it to a line or two.
• Lead with the shot type and one camera move in film terms, like dolly in or crane up.

Best at

• Full scenes up to 30 seconds at 1080p, with dialogue and sound
• Keeping your cast and products true to up to 10 reference photos
• Signs and labels that keep their spelling while the camera moves

Pick another model when

• You want a clip with no soundtrack at all: Kling Video
• You want your character to copy the moves in a real video: Kling Motion Control
• You want the clip timed to your own song or voice track: Seedance 2.5

Tips and quirks

• Write timed beats like "0-3 s" that add up to the full length you picked.
• It cuts between shots by default. Write "one continuous shot, no cuts" for a single take.
• There is no sound switch. End with "no background music, no subtitles" to leave both out.
• Give each reference one sentence: who it is and what to keep. @Image1 tags work.

Best at

• Short cinematic clips from 3 to 15 seconds, with or without sound
• Up to 6 labeled shots in one clip with multi-shot mode
• Bringing a still to life, with an optional end frame to land on

Pick another model when

• You need more than 15 seconds or a cast held from reference photos: Kling 4.0
• You want a character to copy the motion of a real video: Kling Motion Control

Tips and quirks

• Put the camera move in the first few words. "Cinematic" on its own does little.
• Quoted lines can show up as captions. End with "no subtitles, no on-screen text".
• It may add music you did not ask for. Write "no background music, no score" to stop it.
• Keep spoken lines under 10 words with one speaker per clip for the cleanest lip sync.

Best at

• Putting your character into the moves of a real dance or performance clip
• Complex gestures and full-body motion copied from up to 30 seconds of video
• Keeping the original clip's sound under your new character

Pick another model when

• You want to change something inside a clip and keep its setting: Seedance 2.5 Video Edit
• You only have a photo and want to bring it to life: Kling Video

Tips and quirks

• Use one photo of one person. A sheet with several views comes out as a row of clones.
• The result takes the photo's background, not the video's.
• Pick Video orientation for full-body motion up to 30 seconds, Image for camera moves up to 10.
• Pro renders 1080p and Standard renders 720p.

Best at

• Clips up to 20 seconds with sound made in the same pass as the picture
• Continuing or transforming a clip you already have
• Wide 21:9 cinematic shots at up to 1080p

Pick another model when

• You need one take up to 30 seconds: WAN 3.0
• You want to keep a character consistent from reference photos: MiniMax H3

Tips and quirks

• End the prompt with "no music" if you don't want a score added on its own.
• Use one camera move per clip. Stacked moves give confused motion.
• Extensions go up to 15 seconds each. Extend the result again for more.
• Frames and a source clip can't be combined. A start frame sets the shape, so crop it first.

Best at

• Stylized looks in one click: anime, 3D animation, clay, comic or cyberpunk
• One smooth take that travels from your start frame to your end frame
• Clips of any length from 1 to 15 seconds, with optional sound

Pick another model when

• You want a dialogue scene with several speakers: Kling 4.0
• Your character must come from reference photos: Seedance 2.5

Tips and quirks

• Match your words to the style preset. Say cel-shaded with Anime, stop-motion with Clay.
• For a transition, add both a start and an end frame and turn on multi-clip.
• Aim for 30 to 60 words. Long prompts cram too much motion into the clip.
• Describe a look instead of naming a studio or a famous character, like "3D animated".