Can the RTX 5070 12 GB run HunyuanVideo (13B, original)?
Download Q3_K_M (6.2 GB). With the model's working memory it needs about 10.5 GB, leaving 1.5 GB spare on 12 GB. Quality: noticeable loss of detail. For better quality, Q4_K_M (7.9 GB) also runs, with about 0.2 GB spilling into system RAM — a little slower, still practical.
Worth trying on this GPU: with an up-to-date ComfyUI, Dynamic VRAM (on by default for NVIDIA since March 2026) streams whatever does not fit from system RAM, and ComfyUI's own start-up message recommends native FP8/INT8 files over GGUF, saying they “will be faster even if they are larger than your memory”. So before settling for a heavily compressed GGUF, try the native FP8 file (13.2 GB). The file recommended above is the best one that fits entirely — the safe choice on older ComfyUI versions, AMD and Intel. Comfy blog: Dynamic VRAM →
01What to download
| Part | File | Folder | Size | |
|---|---|---|---|---|
| Model | hunyuan-video-t2v-720p-Q3_K_M.gguf Q3_K_M · city96/HunyuanVideo-gguf | models/unet | 6.2 GB | Download → |
| Text encoder | llava_llama3_fp8_scaled.safetensors LLaVA-Llama3 FP8 (scaled) · Comfy-Org/HunyuanVideo_repackaged | models/text_encoders | 9.1 GB | Download → |
| Text encoder | clip_l.safetensors CLIP-L · Comfy-Org/HunyuanVideo_repackaged | models/text_encoders | 0.2 GB | Download → |
| VAE | hunyuan_video_vae_bf16.safetensors HunyuanVideo VAE BF16 · Comfy-Org/HunyuanVideo_repackaged | models/vae | 0.5 GB | Download → |
| Total download · keep about the same free on disk | 16.1 GB | |||
Alternatives: LLaVA-Llama3 FP16 (16.1 GB). The 16-bit text encoder is a little more faithful; the smaller one is chosen here because it loads faster and needs less RAM. Only for some workflows: LLaVA-Llama3 vision (0.6 GB, I2V only). Sizes read from Hugging Face (2026-09-25). “Download” links start the file directly; the file name links are the same files the official ComfyUI workflows use.
System RAM: 32 GB or more. ComfyUI keeps the model file, the text encoder and the VAE in system RAM and moves them to the GPU as needed. With this set of files that is about 16.1 GB, plus roughly 6 GB for Windows, ComfyUI and a browser: 22.1 GB in total. With less RAM it still runs, but Windows starts swapping to disk and loading gets very slow. calculated
02Every HunyuanVideo 13B file on 12 GB
| File | Size | Needed | On this card | Quality | Download |
|---|---|---|---|---|---|
| 16-bit SAFETENSORS · Comfy-Org | 25.6 GB | 29.9 GB | 17.9 GB too big | the original weights | Hugging Face → |
| FP8 SAFETENSORS · Kijai | 13.2 GB | 17.5 GB | 5.5 GB too big | practically identical to the original | Hugging Face → |
| Q8_0 GGUF · city96 | 14.0 GB | 18.3 GB | 6.3 GB too big | practically identical to the original | Hugging Face → |
| Q6_K GGUF · city96 | 11.0 GB | 15.3 GB | 3.3 GB too big | very close to the original | Hugging Face → |
| Q5_K_M GGUF · city96 | 9.4 GB | 13.7 GB | spills 1.7 GB | close; small differences in fine detail | Hugging Face → |
| Q4_K_M GGUF · city96 | 7.9 GB | 12.2 GB | spills 0.2 GB | good; some loss in fine detail and text | Hugging Face → |
| Q3_K_M ← GGUF · city96 | 6.2 GB | 10.5 GB | fits · 1.5 GB spare | noticeable loss of detail | Hugging Face → |
Sizes from the Hugging Face file listing, read 2026-09-25. “Needed” = file + 3.5 GB working memory for this model + 0.8 GB kept free for the system.
03The text encoder
LLaVA-LLaMA-3 8B + CLIP-L: 16.1 GB as 16-bit, 9.1 GB as FP8 (plus CLIP-L (0.25 GB)). ComfyUI encodes the prompt first and can push the encoder out of VRAM before sampling, so it does not have to fit together with the model. On 12 GB the FP8 encoder fits on its own, so prompt encoding stays fast.
04About this GPU
The RTX 5070 12 GB is a Blackwell GPU with FP8 and FP4 hardware: ComfyUI computes Comfy-Org's FP8 files natively here, and NVFP4 files (where a model offers them) are faster still.
RTX 5070 12 GB: 12 GB GDDR7 · 192-bit · 672 GB/s · Blackwell. Everything that runs on the RTX 5070 12 GB →
05What more VRAM would change
With 20 GB you could run HunyuanVideo 13B at 8-bit or better (FP8, 13.2 GB) with no offloading.
What changes from the RTX 5070 12 GB to the RTX 5070 Ti 16 GB →
06Measured and reported results
Nobody has sent measured numbers for this pair yet. If you run HunyuanVideo 13B on a RTX 5070 12 GB, send your time per image and peak VRAM and it will appear here, credited.
07Questions
How much VRAM does HunyuanVideo 13B need?
Around 10.5 GB with the smallest file (Q3_K_M) and 17.5 GB with an 8-bit file (FP8), counting working memory and a small system reserve. The full 16-bit file needs about 29.9 GB.
Which HunyuanVideo 13B file should I download for the RTX 5070 12 GB?
Q3_K_M (6.2 GB) from city96/HunyuanVideo-gguf. Load it with Unet Loader (GGUF) from the ComfyUI-GGUF node pack.
Is FP8 faster than GGUF on the RTX 5070 12 GB?
It can be. This GPU has FP8 hardware, and ComfyUI computes FP8 natively for files made for it (Comfy-Org's fp8_scaled files), or for any FP8 file with the --fast fp8_matrix_mult option. GGUF files are unpacked on the fly, which costs some speed.
How much do I need to download for HunyuanVideo 13B on the RTX 5070 12 GB?
About 16.1 GB for the model file, text encoder and VAE listed on this page, and the same again free on disk. With a 12 GB card, 32 GB of system RAM or more is recommended.