iris-3b by Henry Lee: Open-Weight Model

Model Card

By Henry Lee, published under apache-2.0, revision 1a49f7b495f1.

ComfyUI-ready files for Iris-3B by Speridlabs, a 3B text-to-image transformer that works directly in pixel space (no VAE). Use them with ComfyUI-Iris. It adds native Iris support without any custom nodes and has an example workflow for each file below. qwen3vl4bbf16.safetensors from Comfy-Org/Qwen3-VL. Load it with Load CLIP, type krea2. Text to image. Load Diffusion Model, Load CLIP (krea2), Load VAE (pixelspace), Empty Latent Image at about 1 megapixel, and an empty negative prompt. resmultistep or euler, simple, 50 steps, CFG 3. Upscale x4. Base model plus the upscaler LoRA. Scale the image up 4x with bicubic, VAE Encode, multiply the latent by 2 (LatentMultiply), then one…

Read Henry Lee's full model card

Iris-3B for ComfyUI

ComfyUI-ready files for Iris-3B by Speridlabs, a 3B text-to-image transformer that works directly in pixel space (no VAE).

Use them with ComfyUI-Iris. It adds native Iris support without any custom nodes and has an example workflow for each file below.

Files

Text to image

File	Size	Notes
`iris-3b_bf16.safetensors`	6.0 GB	Recommended.
`iris-3b_int8_convrot.safetensors`	3.4 GB	About 1.7x faster. Needs an RTX 20-series or newer.
`iris-3b.safetensors`	12 GB	The original FP32 release, unchanged.

Depth

File	Size	Notes
`iris-3b-depth_bf16.safetensors`	6.0 GB	Iris-3B's monocular depth fine-tune.
`iris-3b-depth_int8_convrot.safetensors`	3.4 GB	int8 version.

Upscaling

File	Size	Notes
`loras/iris-3b-upscaler_r128.safetensors`	0.5 GB	Iris-3B's restoration fine-tune as a LoRA for the base model.

Text encoder

qwen3vl_4b_bf16.safetensors from Comfy-Org/Qwen3-VL. Load it with Load CLIP, type krea2.

Using them

Text to image. Load Diffusion Model, Load CLIP (krea2), Load VAE (pixel_space), Empty Latent Image at about 1 megapixel, and an empty negative prompt. res_multistep or euler, simple, 50 steps, CFG 3.

Upscale x4. Base model plus the upscaler LoRA. Scale the image up 4x with bicubic, VAE Encode, multiply the latent by 2 (LatentMultiply), then one KSamplerAdvanced step: add_noise disable, steps 2, start at step 1, CFG 1, with ModelSamplingSD3 shift 1. Inputs of 256 to 384 px work best.

Depth. Scale the image to roughly 0.6 to 1 megapixel, VAE Encode, then one KSamplerAdvanced step with add_noise disabled and CFG 1. Small inputs (under about 600 px) come out blocky, so scale them up first. The output is relative depth: dark is near, bright is far.

How they were made

bf16: a direct cast of the FP32 release.
int8 convrot: the trunk MLP width (6826) is zero-padded to 6912 so every layer can be quantized. The padding doesn't change the output. Quantized with comfy-model-tools' quant_int8_convrot.py (mean weight error 0.85%). ComfyUI-Iris pads every checkpoint the same way when loading, so one LoRA fits all versions.
Depth: the released depth model takes an extra input channel that is always zero, plus a small output layer that turns three channels into one. Both are merged into the standard Iris layout without changing the result (maximum difference 0.00014 in FP32). The file is tagged so ComfyUI-Iris knows to output the depth map directly.
Upscaler LoRA: the difference between the restoration fine-tune and the base model, compressed to rank 128 with SVD. Norms, biases and embeddings are stored exactly. Its output matches the full fine-tune at about 37 dB PSNR.

The conversion scripts are in the scripts/ folder of ComfyUI-Iris.

License

Apache-2.0, same as the original. The model and both fine-tunes are by Speridlabs (paper, code).

Identity and Version

Repository: nynxz/iris-3b
Publisher: Henry Lee
Task: Depth estimation
Modality: Other
Library: Not stated by the source
Parameters: Not stated by the source
Languages: Not stated by the source
Revision: 1a49f7b495f1a99e41a898d94caccbcf74d8414c
First published: 2026-10-09
Last updated: 2026-10-10

Files and Weights

8 files, 31.1 GB in total. The weights are 6 files totalling 31.1 GB in safetensors.

Weights6 files · 31.1 GB

Documentation1 file · 3.4 KB

Repository1 file · 1.5 KB

Every file

File	Type	Size	SHA-256
iris-3b-depth_bf16.safetensors	Weights	6.0 GB	12a592b72bc4
iris-3b-depth_int8_convrot.safetensors	Weights	3.4 GB	0be4b7d0ae56
iris-3b.safetensors	Weights	12.0 GB	a90e1f4c197b
iris-3b_bf16.safetensors	Weights	6.0 GB	370cf2235797
iris-3b_int8_convrot.safetensors	Weights	3.4 GB	6ec133c49050
loras/iris-3b-upscaler_r128.safetensors	Weights	502.7 MB	215859e0ee85
README.md	Documentation	3.4 KB	—
.gitattributes	Repository	1.5 KB	—

License and Download

License: apache-2.0
Access: Open weights, no gate
Download size: 31.1 GB

Download from Henry Lee

Released by Henry Lee through its official repository on Hugging Face. Read the license.

Built From

Derived from speridlabs/iris-3b
Described by arXiv:2610.09450

Memory Requirements

Precision	Weights in memory
As published	31.1 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.

Questions About iris-3b

Can I use iris-3b commercially?

Yes. iris-3b is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.

Similar Models

Model · Depth estimation

dpt-hybrid-midas

Intel

Dense Prediction Transformer (DPT) model trained on 1.4 million images for monocular depth estimation. It was introduced in the paper Vision Transformers for Dense Prediction by Ranftl et al. (2021) and first released in this repository. DPT uses the Vision Transformer (ViT) as backbone and adds a neck + head on top for monocular depth estimation. This repository hosts the "hybrid" version of the model as stated in the paper. DPT-Hybrid diverges from DPT by using ViT-hybrid as a backbone and taking some activations from the backbone. The model card has been written in combination by the Hugging Face team and Intel. Here is how to use this model for zero-shot depth estimation on an image…

Open weights apache-2.0 transformers

View model

Model · Depth estimation

Depth-Anything-3-Small-LiteRT

LiteRT Community (FKA TFLite)

On-device LiteRT / TFLite conversion of Depth Anything 3 — Small (ByteDance-Seed, Apache-2.0) for monocular depth, running fully on the mobile GPU via the LiteRT CompiledModel API (ML Drift delegate). No CPU fallback ops — the whole graph is GPU-compatible. DA3 processes images at their native aspect ratio (upperboundresize, longer side → 896, multiple of 14). Forcing a square 896×896 and letterbox-padding drops the match to corr 0.977 (the black padding leaks into the content through global attention). Converting at the native rectangle restores corr 0.9994 and is also faster (fewer tokens). This checkpoint is built for portrait ~9:16. For another aspect, re-convert at that shape (or your…

Open weights apache-2.0 litert

View model

Model · Depth estimation

RK182X-DEPTH-DA3-BASE

RKNNAI

本仓库提供由 depth-anything/DA3-BASE 转换的 RKNN 模型。 - Model ID：RKNNAI/RK182X-DEPTH-DA3-BASE - 模型显示名称：RK182X-DEPTH-DA3-BASE - 源模型：depth-anything/DA3-BASE - 模型类型：DEPTH ModelScope 完整下载： Hugging Face 完整下载： ModelScope 指定配置下载： Hugging Face 指定配置下载： - 使用配套 RKNN Runtime 和驱动；运行前用 rknn-smi -v 检查设备端版本。 - 源模型许可证：Apache License 2.0，正文见 LICENSE，归属与转换修改说明见 NOTICE。 - 同时遵守 RKNN Toolkit、RKNN Runtime 相关许可；本包不包含运行库。

Open weights apache-2.0

View model

Model · Depth estimation

GRADE

Physical Intelligence Rice

This repository contains the complete GRADE model, paper ablations, and baselines: 19 model checkpoint files plus one auxiliary TAESD weight file (20.safetensors files in total). It also includes the inference and metric code and the saved camera-ready quantitative results. The full GRADE model uses the radar, diffusion, and ControlNet checkpoints in checkpoints/grade/. - E1 — saved-result reproduction: regenerate the quantitative tables and figures from the supplied camera-ready CSVs and auxiliary evaluation data. This requires no inference or dataset download. - E2 — local reproduction: run inference on the released evaluation data, compute 2D/3D metrics, then regenerate the quantitative…

Open weights cc-by-nc-4.0 pytorch

View model

Model · Depth estimation

SemiSupMDE

Amir El-Ghoussani

Models for the paper Consistency Regularisation for Unsupervised Domain Adaptation in Monocular Depth Estimation (CoLLAs 2024) by Amir El-Ghoussani, Julia Hornauer, Gustavo Carneiro and Vasileios Belagiannis. Code: github.com/AmirMaEl/SemiSupMDE. A single U-Net (54.6M parameters) predicts depth from one RGB image. It is trained on synthetic Virtual KITTI with ground truth and adapted to real KITTI without any KITTI labels, using a pairwise loss on source images and a perturbation-consistency loss across RandAugment views of the unlabelled target images. Both checkpoints are PyTorch state dicts. KITTI Eigen test split (697 images), Garg crop, velodyne ground truth, no median scaling: With…

Open weights pytorch

View model

Model · Depth estimation

Depth-Anything-V2-Small-hf

Depth Aything

Depth Anything V2 is trained from 595K synthetic labeled images and 62M+ real unlabeled images, providing the most capable monocular depth estimation (MDE) model with the following features: - more fine-grained details than Depth Anything V1 - more robust than Depth Anything V1 and SD-based models (e.g., Marigold, Geowizard) - more efficient (10x faster) and more lightweight than SD-based models - impressive fine-tuned performance with our pre-trained models This model checkpoint is compatible with the transformers library. Depth Anything V2 was introduced in the paper of the same name by Lihe Yang et al. It uses the same architecture as the original Depth Anything release, but uses…

Open weights apache-2.0 25M parameters transformers

View model