Luma UNI 1 Prompting Guide
Luma UNI 1 does both text-to-image and reference-guided editing from one prompt box, in two tiers: Standard at 1 credit for fast iteration, and MAX at 3 credits for the quality-first variant. It ships ~3K output across seven aspect ratios, takes up to eight reference images for edits, and has an optional manga style preset.
Luma UNI 1 is a unified generate-and-edit model. With no reference image it's text-to-image; add one or more references and it switches to edit mode, restyling or recomposing them while following your prompt. You pick the tier per generation — Standard (1 credit) is fast and clean, MAX (3 credits) pushes detail and fidelity.
Output lands around 3K across 1:1, 16:9, 9:16, 3:2, 2:3, 2:1, and 1:2 — wide enough for print crops without a separate upscale. The optional manga preset biases the look toward inked panels; otherwise the model stays photoreal or follows whatever medium you describe.
2
Tiers — Standard · MAX
~3K
Native resolution
8
Reference images for edits
No reference image means text-to-image. Add one and Luma switches to edit mode — restyling, recomposing, or extending the reference while following your prompt. Same prompt box, two workflows.
MAX is the variant to reach for on hero shots — dense concept scenes, photoreal portraits, anything where fidelity matters. Standard handles the bulk of iteration at a third of the cost; switch up only when the frame is a keeper.
Every generation lands around 3K across square, widescreen, vertical, and the ultrawide 2:1 / 1:2 ratios — enough resolution for print crops and large displays without a separate upscale pass.
Flip the style to manga and Luma biases toward crisp ink lines and screentone shading. It's a one-switch shortcut to the look — available on text-to-image (the preset is off when you supply reference images).
Each example shows the exact prompt that produced the result. Copy any prompt with one click.
MAX · 16:9 · 3 credits
A sweeping alien archipelago at sunset, bioluminescent turquoise water between jagged islands, distant twin moons, cinematic concept art, volumetric light, ultra-detailed
MAX earns its credits on busy scenes — atmospheric light, lots of small detail, wide vistas. Stack the scene cues (bioluminescence, twin moons, volumetric light) and let MAX resolve them cleanly.
MAX · 2:3 · 3 credits
Photoreal portrait of a weathered elderly fisherman, deep wrinkles, knit cap, golden-hour light, documentary photography, sharp catchlight in the eyes, shallow depth of field
For faces, MAX holds skin texture and micro-detail that Standard softens. Documentary cues ("golden-hour light", "sharp catchlight", "shallow depth of field") give it a clear photographic target.
MAX · 2:3 · reference image · 3 credits
Restyle this portrait as a 1970s film still — warm Kodachrome grain, soft directional window light, muted earth tones, vintage cinematic color, keep the subject identity
Supply a portrait as a reference and describe only the change — film era, grain, color, light. "Keep the subject identity" tells Luma to restyle rather than reinvent the person.
Standard · 2:3 · manga style · 1 credit
A lone swordsman standing on a rooftop at dusk, wind catching his long coat, dramatic clouds behind, dynamic panel composition, crisp ink lines and screentone shading
Turn the style preset to manga and describe composition like a panel. Standard is plenty for stylized line work — you don't need MAX when the look is graphic rather than photoreal.
Standard (1 credit) is fast and clean — use it to find the composition and wording. Switch to MAX (3 credits) only for the final frame, dense scenes, or photoreal faces where the extra fidelity shows.
With no reference, Luma generates from scratch. Add one (or up to eight) and it edits — restyle, recompose, extend. In edit mode, describe the change, not the whole scene, so the model keeps what you want from the source.
The manga style switch is the fastest route to crisp ink-and-screentone panels. It applies to text-to-image; when you supply reference images the preset turns off, so style those edits through the prompt instead.
Output is already around 3K, so you can crop in or send to print without a separate upscale for most uses. Pick the aspect ratio to match the final placement — including 2:1 / 1:2 for ultrawide banners and tall posters.
You can pass up to eight references, but conflicting styles pull the result in different directions. Group references that share a look or purpose, and name in the prompt what each should contribute.
When editing a portrait, an explicit identity cue ("keep the subject identity / same face") biases Luma toward restyling rather than redrawing the person — important for character continuity.
| Setting | Values | Notes |
|---|---|---|
| Model variant | Standard · MAX | Standard = 1 credit (fast). MAX = 3 credits (quality-first). |
| Mode | Text-to-image · Edit | Adding reference images switches to edit mode automatically. |
| Aspect ratio | 1:1 · 16:9 · 9:16 · 3:2 · 2:3 · 2:1 · 1:2 | Seven presets resolving to ~3K dimensions. |
| Reference images | Up to 8 | Edit / guidance source. More refs = more to reconcile. |
| Style | auto · manga | manga biases toward inked panels; available on text-to-image only. |
| Number of outputs | 1–4 | Variations per call; multiplies credit cost. |
| Pricing | Per image | Standard 1 credit, MAX 3 credits. |
Standard (1 credit) is the fast, clean tier for iteration and most everyday work. MAX (3 credits) is quality-first — it resolves more detail on dense scenes and photoreal faces. Iterate on Standard, finalize on MAX when the extra fidelity is worth it.
Luma UNI 1 is unified: with no reference image it generates from your prompt; add one or more reference images and it switches to edit mode, restyling or recomposing them while following the prompt. Describe the change rather than the whole scene for the cleanest edits.
It biases the output toward crisp ink lines and screentone shading — the manga/comic panel look — without you having to describe it in detail. It applies to text-to-image generations; when you supply reference images the preset is off, so steer the style through your prompt.
Around 3K across seven aspect ratios (1:1, 16:9, 9:16, 3:2, 2:3, 2:1, 1:2). That’s enough for print crops and large displays in most cases without a separate upscale step.
Up to eight. They drive edit mode and guidance — grouping references that share a look gives more predictable results than mixing conflicting styles.
Yes. POST to /api/v1/models/luma-uni-1/run with prompt, model (standard/max), aspect_ratio, num_outputs, optional style, and reference_images (which switches to edit mode). The output is returned as image URLs.