You need to agree to share your Hugging Face information to access this model
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
These files are Derivatives of LTX-2.x and are subject to the LTX-2.x Community License Agreement (LICENSE.txt), including Section 4 and Attachment A (use restrictions and acceptable use policy). Commercial Entities (annual revenues of at least USD 10,000,000) must obtain a paid license from Lightricks (licensing@ltx.io) before any use other than a Non-Commercial Purpose (Section 2.2).
Log in or Sign Up to review the conditions and access this model content.
LTX-2.5 22B dev — ROCmFP4 GGUF
The full (non-distilled) video transformer of Lightricks/LTX-2.5,
quantized to Q4_0_ROCMFP4, the 4-bit ROCmFPX format for AMD GPUs, for
stable-diffusion.cpp. The dev model is the one for full-quality
sampling with CFG and for LoRA work; for fast 8-step generation see
Harsha001/LTX-2.5-22B-distilled-ROCmFPX.
A version with an uncensored text encoder is at
Harsha001/LTX-2.5-22B-dev-uncensored-ROCmFPX.
This file needs stable-diffusion.cpp built against the ROCmFPX fork of ggml. Stock sd.cpp, ComfyUI-GGUF and other runners do not know this type. The patch is in this repo (
sdcpp-rocmfpx.patch).Not an official Lightricks release. Use is subject to the LTX-2.x Community License Agreement (
LICENSE.txt,NOTICE.md): Commercial Entities with annual revenues of at least USD 10 million need a paid license from Lightricks for commercial use.
Measured quality
4 prompts, 768×448, 33 frames, Euler, 20 steps, CFG 3.0, the negative prompt below, same seed per prompt. Every frame is compared with the frame of the same prompt, seed and index from the reference (132 frames per row). PSNR (dB) and SSIM: higher is closer. LPIPS: lower is closer (perceptual distance). Diffusion amplifies small weight changes over the steps, so a quantized model can give a different take of the same prompt (pose, composition) without any loss of quality. The reference is the BF16 model.
| File | Size | PSNR | SSIM | LPIPS |
|---|---|---|---|---|
| Same BF16 weights loaded directly (the floor of this test) | 42.0 GB | 41.1 | 0.988 | 0.015 |
LTX-2.5-22B-dev-Q4_0_ROCMFP4.gguf |
14.7 GB | 24.6 | 0.871 | 0.147 |
| Q4_K (stock ggml, same norm rule, baseline only) | 14.7 GB | 26.1 | 0.888 | 0.126 |
Honest comparison: stock Q4_K of about the same size (14.7 GB) stays closer to BF16 on this test. ROCmFP4 is built for fast AMD kernels; its speed for video models on AMD has not been measured yet.
Norm weights stay in BF16. stable-diffusion.cpp's converter quantizes every tensor whose length fits the
block size, including the 1-D RMSNorm weights. Those hold outliers (in LTX-2.5's attention q/k norms one value per
32-value block can be 67 times the block median; in the Gemma encoder's layer norms values reach 600 next to a median
of 7), so in 4- or 8-bit blocks most of their values are lost. All files here keep them in BF16 with
--tensor-type-rules 'norm\.weight=bf16' (a few MB).
The first upload of this file (20 October 2026) still had quantized norm weights and scored PSNR 23.5, SSIM 0.842, LPIPS 0.177 on the same test; it was replaced the same day.
Sizes are the diffusion transformer only. Biases, norms, scale/shift tables and the keyframe position embedding stay unquantized.
How this was made
- stable-diffusion.cpp (
b13a482) built on the ROCmFPX ggml (fb08d7c) withsdcpp-rocmfpx.patch. - Importance matrix from the BF16 dev model over 6 calibration prompts (different from the test prompts), same settings.
sd-cli -M convert --type q4_0_rocmfp4 --tensor-type-rules 'keyframes_abs_pos_embedding=f32,norm\.weight=bf16' --imat-in imatrix.dat(the norm rule: see the note on norm weights).- Quality measured on GPU with every quantized weight expanded back to full precision and rendered with stock sd.cpp, so the numbers measure the stored weights. The fork's own GPU kernels for these types were not used (they are wrong on NVIDIA). AMD GPU kernels compute the same math in a different order, so results on an AMD GPU can differ slightly. These files have not been tested on AMD hardware yet.
Run
git clone https://github.com/charlie12345/ROCmFPX
git clone https://github.com/leejet/stable-diffusion.cpp sdcpp && cd sdcpp && git checkout b13a482
git submodule update --init --recursive && git apply ../sdcpp-rocmfpx.patch
cmake -B build -DSD_USE_UPSTREAM_GGML=ON -DSD_GGML_SOURCE_DIR=$PWD/../ROCmFPX/ggml -DSD_ROCMFPX=ON -DSD_VULKAN=ON
cmake --build build -j
build/bin/sd-cli -M vid_gen --diffusion-model LTX-2.5-22B-dev-Q4_0_ROCMFP4.gguf \
--llm gemma4-12b-with-proj-ltx-2.5-bf16.safetensors --vae ltx-2.5-video-vae-conv-bf16.safetensors \
--flash-attn off --cfg-scale 3.0 --steps 20 --sampling-method euler -n "worst quality, low quality, blurry, distorted, artifacts" \
-W 768 -H 448 --video-frames 33 --fps 24 -p "a red fox running through fresh snow" -o fox.webm
Text encoder: text_encoders/gemma4-12b-with-proj-ltx-2.5-bf16.safetensors from
Lightricks/LTX-2.5.
--flash-attn off: on NVIDIA, sd.cpp's flash attention turned every latent into NaN (white frames) with LTX-2.5;
it has not been checked on AMD. The default diffusion VAE decoder is not supported by sd.cpp; use the conv one
from Lightricks/LTX-2.5 (vae/).
- Downloads last month
- -
4-bit
Model tree for Harsha001/LTX-2.5-22B-dev-ROCmFPX
Base model
Lightricks/LTX-2.5