speridlabs/iris-3b — new model trending #30 on Hugging Face
Sperid Labs released Iris-3B, a 3-billion-parameter pixel-space image model that also does depth and upscaling.
Sperid Labs released Iris-3B, a 3-billion-parameter Apache 2.0 model that generates images directly in pixel space without a VAE or latent decoder. Prompts produce roughly 1024-pixel images, with defaults of CFG scale 3 and 100 denoising steps, using Qwen3-VL-4B-Instruct as the text encoder. The same backbone, fine-tuned with no architectural change, is also released for monocular depth estimation and image restoration and upscaling. Text-to-image weights are about 12 GB, and the depth and upscaler folders add about 12 GB each; a paper is listed as arXiv 2610.09450.
- 3B-parameter pixel-space generator with no VAE or latent decoder.
- Default generation uses CFG 3, 100 steps, and about 1024-pixel images.
- Text encoder is Qwen3-VL-4B-Instruct; text-to-image weights are about 12 GB.
- Fine-tunes cover monocular depth, restoration, and upscaling.
- Apache 2.0 release, trending number 30 on Hugging Face.
- Organizations
- Sperid Labs
- AI models
- Iris-3BQwen3-VL-4B-Instruct
From the original article
5 min read at huggingface.cotext-to-image · 393 downloads · 135 likes · license apache-2. 0
By speridlabs · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.