iris-3b
Iris-3B: pixel-space generation & general vision learner. A 3B diffusion transformer that generates every pixel directly (no VAE); the same prior, fine-tuned, does monocular depth estimation and 4x image restoration. Paper, weights, training code and demo.
Get picks like this daily. The day's top launches, AI/tech news, and a weekly opportunity spotlight — straight to your inbox.
This is 1 of 140 launches in open-weight LLM inference and deployment — see how it stacks up on momentum and crowding →
41 other launches read as similar to this one →
Details
- External ID
- 1405931849
- Source
- GITHUB
- Company
- —
- Product
- iris-3b
- Website domain
- speridlabs.com
- Launched
- Oct. 5, 2026
- Cohort
- —
- Upvotes
- 74
- Upvotes percentile
- 0.78
- Tags
- depth-estimation, diffusion, flow-matching, generative-prior, image-restoration, pixel-space, pytorch, super-resolution, text-to-image, vision-foundation-model
Enrichment
- Niche
- open-weight LLM inference and deployment
- Vertical
- Media & entertainment
- Function
- Model & infra
- Audience
- Developer
- AI stance
- AI-native
- Project type
- Hobby / open-source project
- Normalized one-liner
- pixel-space diffusion transformer model for generation and restoration
- Manually corrected
- False
Could you build this?
No Iris-3B is a 3-billion-parameter foundation model operating directly in pixel space without a VAE, requiring novel machine learning research, massive compute clusters, and proprietary training datasets.
What it would actually take: Training a 3B diffusion transformer in pixel space requires millions of dollars in GPU compute (e.g., hundreds of H100s), customized distributed training frameworks (PyTorch, Megatron-LM/DeepSpeed), and novel spatial architectures that handle raw pixel tokens without blowing up attention memory costs. Deep ML research expertise in diffusion architectures and vision representation learning is required.
Competitors
Other products that read as similar to this one — 41 launches clear the similarity bar, closest 8 shown.
Attention rank: #9 of 42 (itself plus its competitors, highest first — normalized so YC and Product Hunt are compared fairly).
Launched 236 days after the earliest competitor.
- iris-3b-metal · github · 2026-10-09 · 45 upvotes · similarity 0.50
- Iris-3B-Comfy-Nodes · github · 2026-10-10 · 12 upvotes · similarity 0.49
- Parallax Iris · ph · 2026-10-10 · 1 upvotes · similarity 0.41
- transformer-architecture · github · 2026-09-11 · 67 upvotes · similarity 0.40
- deepseek-vision-pro · github · 2026-10-13 · 21 upvotes · similarity 0.38
- OptiLens 3D · ph · 2026-09-24 · 1 upvotes · similarity 0.38
- BLINDER · ph · 2026-09-07 · 3 upvotes · similarity 0.37
- Iris Analyzer · ph · 2026-09-27 · 1 upvotes · similarity 0.36
Other launches for this product
- No other launches for this product.