Published: August 6, 2026
Image Model — text-to-image, editing & remix
Reve 2.1 is Reve's text-to-image model, released by the independent Palo Alto lab. It generates, edits, and remixes images natively at 4K, reasoning about layout, hierarchy, and spatial relationships before it renders rather than painting straight from the prompt.
Key Features
Generation Modes
- Text-to-Image: Generate images from a prompt with strong adherence and layout planning.
- Image Editing: Change one element of a source image via a natural-language instruction, without re-rolling the rest.
- Remix: Combine up to 8 reference images into a new composition, blending subjects, styles, or backgrounds.
Layout Intelligence
- Structured layout planning: Reasons about structure, hierarchy, and spatial relationships before rendering, keeping dense or busy compositions coherent.
- Built like code: Treats an image as hierarchical, structured regions rather than raw pixels, making iteration precise and predictable.
Resolution & Detail
- Native 4K output: Every generation renders at a true 4096×4096 px (16 megapixels), with high resolution as a first-class primitive rather than a post-process upscale.
- Print-ready: No separate upscaler in the loop; output is ready for large-format use out of the box.
Text & Typography
- Legible in-image text: Words are placed as part of the layout, so text stays legible even in dense infographics, packaging, and signage.
- Foreign-script rendering: Notably improved multilingual text rendering compared to Reve 2.0.
Technical Capabilities
Details |
|
|---|---|
| Provider | Reve (independent foundation model lab), served via the fal.ai API · commercial use OK |
| Output resolution | Native 4K class — true 16 megapixels per image; 1:1 renders at 4096×4096 px; other aspect ratios scale proportionally at the same ~16MP class |
| Aspect ratio | 4:1, 3:1, 21:9, 2:1, 17:9, 16:9, 3:2, 4:3, 5:4, 1:1, 4:5, 3:4, 2:3, 9:16, 1:2, 1:3, 1:4, or Auto (default) |
| Reference images (Remix) | 1–8 images per request, referenced in the prompt as <frame>N</frame> |
| Input image size (Remix) | Under 10 MB per image |
| Input formats (Remix) | PNG, JPEG, WebP, AVIF, HEIF |
| Output formats | PNG (default), JPEG, WebP |
Limitations
- Content policy is more locked-down than some competitors, which can restrict certain prompts.
Prompting Tips
- In Remix, assign each reference image a clear role in the prompt — subject, style, or background.
- Lean on the model's layout planning for dense compositions like infographics, posters, and packaging.
- For legible in-image text, spell out the exact copy you want rendered, including any foreign-script text.
- Use Edit for single-element changes instead of re-generating the whole image from scratch.