Pixel-space generation & general vision learner: text-to-image, monocular depth, image restoration and upscaling from one 3B pixel-space prior. Paper: arXiv:2610.09450.
Loads exported weights (model.safetensors + config.yaml, from
scripts/export_checkpoint.py) from the model repo in IRIS_MODEL_REPO
(default speridlabs/iris-3b). Set the HF_TOKEN secret while that repo is private.
iris3b/ is a copy of src/iris3b from the Iris code repo.
examples/ holds curated Iris-3B samples shown as presets; examples/presets.json records the
prompt, size, seed, steps and guidance each was sampled with.
Pixel-space generation & general vision learner: text-to-image, monocular depth, image restoration and upscaling from one 3B pixel-space prior. Paper: arXiv:2610.09450.
Loads exported weights (model.safetensors + config.yaml, from
scripts/export_checkpoint.py) from the model repo in IRIS_MODEL_REPO
(default speridlabs/iris-3b). Set the HF_TOKEN secret while that repo is private.
iris3b/ is a copy of src/iris3b from the Iris code repo.
examples/ holds curated Iris-3B samples shown as presets; examples/presets.json records the
prompt, size, seed, steps and guidance each was sampled with.
