HiDream-I1 (Dev) on 8 GB of VRAM
Not entirely. Even Q2_K is 1.4 GB over 8 GB, so ComfyUI keeps part of it in system RAM. A 17B mixture-of-experts image model with four text encoders. The Full version has the same file sizes but uses more steps.
Nothing fits entirely, but Q2_K (6.6 GB) overflows by only about 1.4 GB. ComfyUI keeps that part in system RAM automatically: slower than a full fit, but usable.
Also try the FP8 file (17.1 GB) on NVIDIA GPUs. It does not fit entirely, but a current ComfyUI streams the overflow from RAM and native 8-bit files often beat a small GGUF that does fit. Dynamic VRAM →
01Other amounts of VRAM
02Every file, on 8 GB
| File | Size | Needed | On this card | Quality | Download |
|---|---|---|---|---|---|
| 16-bit SAFETENSORS · Comfy-Org | 34.2 GB | 37.0 GB | 29.0 GB too big | the original weights | Hugging Face → |
| Q8_0 GGUF · city96 | 18.7 GB | 21.5 GB | 13.5 GB too big | practically identical to the original | Hugging Face → |
| FP8 SAFETENSORS · Comfy-Org | 17.1 GB | 19.9 GB | 11.9 GB too big | practically identical to the original | Hugging Face → |
| Q6_K GGUF · city96 | 14.7 GB | 17.5 GB | 9.5 GB too big | very close to the original | Hugging Face → |
| Q5_K_M GGUF · city96 | 13.0 GB | 15.8 GB | 7.8 GB too big | close; small differences in fine detail | Hugging Face → |
| Q4_K_M GGUF · city96 | 11.5 GB | 14.3 GB | 6.3 GB too big | good; some loss in fine detail and text | Hugging Face → |
| Q3_K_M GGUF · city96 | 8.8 GB | 11.6 GB | 3.6 GB too big | noticeable loss of detail | Hugging Face → |
| Q2_K ← GGUF · city96 | 6.6 GB | 9.4 GB | spills 1.4 GB | heavy loss; a last resort | Hugging Face → |
“Needed” adds 2 GB of working memory for HiDream-I1 and a 0.8 GB reserve. Sizes 2026-10-01.
03What to download
| Part | File | Folder | Size | |
|---|---|---|---|---|
| Model | hidream-i1-dev-Q2_K.gguf Q2_K · city96/HiDream-I1-Dev-gguf | models/unet | 6.6 GB | Download → |
| Text encoder | t5xxl_fp8_e4m3fn_scaled.safetensors T5-XXL FP8 scaled · Comfy-Org/HiDream-I1_ComfyUI | models/text_encoders | 5.2 GB | Download → |
| Text encoder | clip_l_hidream.safetensors CLIP-L (HiDream) · Comfy-Org/HiDream-I1_ComfyUI | models/text_encoders | 0.2 GB | Download → |
| Text encoder | clip_g_hidream.safetensors CLIP-G (HiDream) · Comfy-Org/HiDream-I1_ComfyUI | models/text_encoders | 1.4 GB | Download → |
| Text encoder | llama_3.1_8b_instruct_fp8_scaled.safetensors Llama 3.1 8B Instruct FP8 · Comfy-Org/HiDream-I1_ComfyUI | models/text_encoders | 9.1 GB | Download → |
| VAE | ae.safetensors FLUX.1 VAE (ae) · Comfy-Org/HiDream-I1_ComfyUI | models/vae | 0.3 GB | Download → |
| Total download · keep about the same free on disk | 22.8 GB | |||
Alternatives: T5-XXL FP16 (9.8 GB). Sizes from Hugging Face, 2026-10-01.
04The text encoder
At 15.9 GB even the smallest CLIP-L + CLIP-G + T5-XXL + Llama 3.1 8B file is too big for 8 GB, so it runs from system RAM. That slows down reading the prompt, not the picture. Sizes: FP8 15.9 GB (all four together: CLIP-L 0.25 + CLIP-G 1.39 + T5-XXL FP8 5.16 + Llama 3.1 8B FP8 9.08 GB). Why the encoder does not count against VRAM →