Iris-3B
Iris-3B is a 3-billion-parameter pixel-space model that generates images from text prompts at 1024 pixels. Fine-tuned weights in the repository also support monocular depth estimation and image restoration and upscaling.
What it does
Use it to generate 1024-pixel images from text, estimate relative depth from photos, or restore and upscale images.
- Availability
- Weights, code, and interactive demo are provided in the repo and linked Hugging Face Spaces.
- Pricing model
- Not established
- Deployment
- Run locally; requires an NVIDIA GPU with CUDA.
- Privacy
- Not established
- Licence
- Apache 2.0
Capability
Modality
Access
Deployment
Pricing
Target user
Evidence
6 claims- Iris-3B generates images directly in pixel space without a VAE or latent space. [1]
<p align="center"><b>Pixel-space generation & general vision learner.</b><br>A 3-billion-parameter model that paints every pixel directly — no VAE, no latent space — and, fine-tuned, estimates depth and restores and upscales images.</p>
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - The default generated image output is 1024×1024. [1]
Default output is 1024×1024.
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - Fine-tuned versions estimate monocular depth and restore and upscale images. [1]
The same backbone, fine-tuned per task: Iris-3B estimates monocular depth and restores and upscales images.
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - The repository provides downloadable model weights and code, and links to an interactive demo. [1]
- Local generation requires an NVIDIA GPU with CUDA. [1]
You need an NVIDIA GPU with CUDA.
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - Depth outputs are affine-invariant rather than metric. [1]
Sources
1
1
Hugging FaceOwner-controlled
https://huggingface.co/speridlabs/iris-3b
Retrieved 2026-10-29 00:34 UTC
Verified
Versions
3| Version | Effective | Change | Changed fields | State |
|---|---|---|---|---|
| 3 | 2026-10-12 20:03 UTC | material change | deployment, pricing, availability, deployment | Current |
| 2 | 2026-10-11 20:58 UTC | material change | deployment, pricing, availability, deployment | Current |
| 1 | 2026-10-10 21:52 UTC | publication | None | Current |
In briefs
1- Two local models can classify, route, or score text and images Daily brief 2026-10-10
Original catalog content is licensed under CC BY 4.0, attributed to wowsuchai. Source quotations and linked content keep their original rights.