Sperid Labs published Apache-2.0 Iris-3B weights, code, paper and depth/upscaler checkpoints.
Iris-3B explores pixel-space diffusion as a general vision learner
2026-10-05Frontier watchTRIAL
Sperid Labs released Apache-2.0 weights and code for Iris-3B, a 3B pixel-space diffusion model that generates images without a VAE and is fine-tuned for monocular depth and restoration. The model card reports zero-shot depth and 4x restoration results, but evidence remains author-reported and inference requires an NVIDIA GPU plus a separate 4B text encoder for generation.
What this means for you
For vision teams, one backbone spanning generation, depth and restoration is a testable alternative to latent-space pipelines where compression may discard texture. It is still an early research release and should be compared on local dense-vision tasks before architectural conclusions.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Model | 3B diffusion transformer plus frozen Qwen3-VL-4B-Instruct text encoder for generation |
|---|---|
| Method | direct pixel-space rectified-flow generation without a VAE; task-specific fine-tuning without backbone architecture changes |
| Artifacts | Apache-2.0 model weights, depth and upscaler checkpoints, code, paper and public demo |
| Author Results | depth AbsRel 0.071 and delta1 0.946 mean across five datasets; restoration LPIPS 0.292 and PSNR 20.68 on DIV2K validation |
| Limits | about 12 GB per checkpoint family, NVIDIA CUDA required, 100 default denoising steps, author-reported evaluation |
| Time Basis | Hugging Face repository created 2026-10-05T14:51:15Z and last modified 2026-10-08T16:51:50Z; recovered by recall audit |
Technical impact 4
Engineering value 3
Adoption signal 3
Maturity 2
Verification cost 3
Risk 2