HiDream-I1 (Dev) VRAM requirements
A 17B mixture-of-experts image model with four text encoders. The Full version has the same file sizes but uses more steps.
TypeImage
Parameters17B
Fits entirely from10 GB
8-bit or better from20 GB
01The files, and how much VRAM each needs
| File | Size | Needed | Min. VRAM | Quality | Source |
|---|---|---|---|---|---|
| 16-bit | 34.2 GB | 37.0 GB | 38 GB | the original weights | Comfy-Org/HiDream-I1_ComfyUI → |
| Q8_0 | 18.7 GB | 21.5 GB | 22 GB | practically identical to the original | city96/HiDream-I1-Dev-gguf → |
| FP8 | 17.1 GB | 19.9 GB | 20 GB | practically identical to the original | Comfy-Org/HiDream-I1_ComfyUI → |
| Q6_K | 14.7 GB | 17.5 GB | 18 GB | very close to the original | city96/HiDream-I1-Dev-gguf → |
| Q5_K_M | 13.0 GB | 15.8 GB | 16 GB | close; small differences in fine detail | city96/HiDream-I1-Dev-gguf → |
| Q4_K_M | 11.5 GB | 14.3 GB | 15 GB | good; some loss in fine detail and text | city96/HiDream-I1-Dev-gguf → |
| Q3_K_M | 8.8 GB | 11.6 GB | 12 GB | noticeable loss of detail | city96/HiDream-I1-Dev-gguf → |
| Q2_K | 6.6 GB | 9.4 GB | 10 GB | heavy loss; a last resort | city96/HiDream-I1-Dev-gguf → |
“Needed” = file + 2 GB working memory + 0.8 GB system reserve.
02By amount of VRAM
03Best GPU for HiDream-I1
The cheapest cards (by launch price) that run it well, and every card sorted by memory: best GPU for HiDream-I1 → Planning bigger images or longer clips? Open the calculator →
04By graphics card
| GPU | VRAM | Verdict | Best file | Needed |
|---|---|---|---|---|
| Desktop graphics cards | ||||
| RTX 2060 6 GB | 6 GB | Offload only | Q4_K_M | 14.3 GB |
| RTX 3050 6 GB | 6 GB | Offload only | Q4_K_M | 14.3 GB |
| RTX 2070 Super 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 2080 Super 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 3050 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 3060 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 3060 Ti 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 3070 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 3070 Ti 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 4060 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 4060 Ti 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 5050 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 5060 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 5060 Ti 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RX 7600 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RX 9050 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RX 9060 XT 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| Arc B570 10 GB | 10 GB | Tight | Q2_K | 9.4 GB |
| RTX 3080 10 GB | 10 GB | Tight | Q2_K | 9.4 GB |
| RTX 2080 Ti 11 GB | 11 GB | Tight | Q2_K | 9.4 GB |
| Arc B580 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RTX 2060 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RTX 3060 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RTX 3080 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RTX 3080 Ti 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RTX 4070 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RTX 4070 Super 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RTX 4070 Ti 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RTX 5070 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RX 7700 XT 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RX 9070 GRE 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| Arc A770 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RTX 4060 Ti 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RTX 4070 Ti Super 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RTX 4080 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RTX 4080 Super 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RTX 5060 Ti 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RTX 5070 Ti 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RTX 5080 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RX 7600 XT 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RX 7800 XT 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RX 7900 GRE 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RX 9060 XT 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RX 9070 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RX 9070 XT 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RX 7900 XT 20 GB | 20 GB | Runs well | FP8 | 19.9 GB |
| Arc Pro B60 24 GB | 24 GB | Runs well | Q8_0 | 21.5 GB |
| RTX 3090 24 GB | 24 GB | Runs well | Q8_0 | 21.5 GB |
| RTX 3090 Ti 24 GB | 24 GB | Runs well | Q8_0 | 21.5 GB |
| RTX 4090 24 GB | 24 GB | Runs well | FP8 | 19.9 GB |
| RX 7900 XTX 24 GB | 24 GB | Runs well | Q8_0 | 21.5 GB |
| Arc Pro B70 32 GB | 32 GB | Runs well | Q8_0 | 21.5 GB |
| RTX 5090 32 GB | 32 GB | Runs well | FP8 | 19.9 GB |
| Laptop GPUs | ||||
| RTX 3050 Laptop 4 GB | 4 GB | Not practical | Q2_K | 9.4 GB |
| RTX 3050 Ti Laptop 4 GB | 4 GB | Not practical | Q2_K | 9.4 GB |
| RTX 2060 Laptop 6 GB | 6 GB | Offload only | Q4_K_M | 14.3 GB |
| RTX 3050 Laptop 6 GB | 6 GB | Offload only | Q4_K_M | 14.3 GB |
| RTX 3060 Laptop 6 GB | 6 GB | Offload only | Q4_K_M | 14.3 GB |
| RTX 4050 Laptop 6 GB | 6 GB | Offload only | Q4_K_M | 14.3 GB |
| RTX 2070 Laptop 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 2070 Super Laptop 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 2080 Laptop 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 2080 Super Laptop 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 3070 Laptop 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 3070 Ti Laptop 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 3080 Laptop 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 4060 Laptop 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 4070 Laptop 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 5050 Laptop 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 5060 Laptop 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 5070 Laptop 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RX 7600M 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RX 7600M XT 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RX 7600S 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RX 7700S 8 GB | 8 GB | Offload only | Q2_K | 9.4 GB |
| RTX 4080 Laptop 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RTX 5070 Laptop 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RTX 5070 Ti Laptop 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RX 7800M 12 GB | 12 GB | Tight | Q3_K_M | 11.6 GB |
| RTX 3080 Laptop 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RTX 3080 Ti Laptop 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RTX 4090 Laptop 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RTX 5080 Laptop 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RX 7900M 16 GB | 16 GB | Runs | Q5_K_M | 15.8 GB |
| RTX 5090 Laptop 24 GB | 24 GB | Runs well | FP8 | 19.9 GB |
| Unified memory | ||||
| Radeon 8060S (Strix Halo) 96 GB | 96 GB | Runs well | 16-bit | 37.0 GB |
| Radeon 8065S (Gorgon Halo) 160 GB | 160 GB | Runs well | 16-bit | 37.0 GB |
| Apple Silicon Macs (by memory) | ||||
| Mac 16 GB | 12.7 GB | Tight | Q3_K_M | 11.6 GB |
| Mac 18 GB | 14.4 GB | Runs | Q4_K_M | 14.3 GB |
| Mac 24 GB | 19.6 GB | Runs | Q6_K | 17.5 GB |
| Mac 32 GB | 26.8 GB | Runs well | Q8_0 | 21.5 GB |
| Mac 36 GB | 30.2 GB | Runs well | Q8_0 | 21.5 GB |
| Mac 48 GB | 40.2 GB | Runs well | 16-bit | 37.0 GB |
| Mac 64 GB | 55.7 GB | Runs well | 16-bit | 37.0 GB |
| Mac 96 GB | 85 GB | Runs well | 16-bit | 37.0 GB |
| Mac 128 GB | 115.4 GB | Runs well | 16-bit | 37.0 GB |
| Mac 192 GB | 175.4 GB | Runs well | 16-bit | 37.0 GB |
| Mac 256 GB | 236.9 GB | Runs well | 16-bit | 37.0 GB |
| Mac 512 GB | 498.1 GB | Runs well | 16-bit | 37.0 GB |
05Text encoder, VAE and other files
| File | Folder | Size | When | |
|---|---|---|---|---|
| CLIP-L (HiDream) clip_l_hidream.safetensors | models/text_encoders | 0.2 GB | required | Download → |
| CLIP-G (HiDream) clip_g_hidream.safetensors | models/text_encoders | 1.4 GB | required | Download → |
| T5-XXL FP8 scaled t5xxl_fp8_e4m3fn_scaled.safetensors | models/text_encoders | 5.2 GB | text encoder · smaller, recommended | Download → |
| T5-XXL FP16 t5xxl_fp16.safetensors | models/text_encoders | 9.8 GB | text encoder · alternative | Download → |
| Llama 3.1 8B Instruct FP8 llama_3.1_8b_instruct_fp8_scaled.safetensors | models/text_encoders | 9.1 GB | required | Download → |
| FLUX.1 VAE (ae) ae.safetensors | models/vae | 0.3 GB | required | Download → |
The files the official ComfyUI workflows load next to the model. Sizes read from Hugging Face (2026-10-01).
At 15.9 GB even the smallest CLIP-L + CLIP-G + T5-XXL + Llama 3.1 8B file is too big for 16 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 →
06Where the files go in ComfyUI
| File | Folder | Loader node |
|---|---|---|
| Diffusion model (.safetensors: 16-bit, FP8, INT8) | ComfyUI/models/diffusion_models | Load Diffusion Model |
| GGUF file (.gguf) | ComfyUI/models/unet | Unet Loader (GGUF) - from the ComfyUI-GGUF node pack |
| Text encoder | ComfyUI/models/text_encoders | Load CLIP / DualCLIPLoader (or the GGUF versions) |
| VAE | ComfyUI/models/vae | Load VAE |
07Measured and reported results
| Label | GPU | Setup | Result | Peak VRAM | Date | Source |
|---|---|---|---|---|---|---|
| reported | RTX 3090 24 GB | HiDream Full fp8 (T5 fp8, Llama3.1 fp8 scaled) “ComfyUI Full example from their page at https://comfyanonymous.github.io/ComfyUI_examples/hidream/ runs at 115.09s on my RTX 3090, from cold start.” HiDream-I1 Full via ComfyUI example workflow, Windows portable; cold start incl. load; resolution/steps not restated | 115.09 s / image | — | 2025-04-18 | huggingface.co → |
Other people's numbers are copied as published, with a link. Send yours.
08Training a LoRA for HiDream-I1
| Trainer | VRAM | Type | Settings and quote | Source |
|---|---|---|---|---|
| SimpleTuner | 20 GB | stated minimum | Lowest config: int8-quanto or nf4-bnb base, 1024px, batch 1, Lion8bit paged, gradient checkpointing, tiny LoRA (rank 1) or LoKr factor 25; guide also says 24G is likely minimum without offload “The lowest VRAM HiDream configuration is about 20-22G:” | github.com → |
| diffusion-pipe | 24 GB | stated minimum | Needs block swapping or nf4 transformer (possible quality loss); 48GB (or 2x24GB pipeline parallel) without block swap “With enough block swapping you can train on a single 24GB GPU.” | github.com → |
| ai-toolkit | 35.2 GB | example run | Official example config (file named 48): base + TEs quantized, rank 32, batch 1, gradient checkpointing; not possible on single 24GB card per config “The settings here will take ~35.2GB of vram to train.” | github.com → |
reported From each trainer's own docs, 2026-09-25. Why the figures differ →
09Every file tracked for HiDream-I1
| File | Type | Size | Repo |
|---|---|---|---|
| hidream_i1_dev_bf16.safetensors | BF16 | 34.21 GB | Comfy-Org/HiDream-I1_ComfyUI → |
| hidream_i1_dev_fp8.safetensors | FP8 | 17.11 GB | Comfy-Org/HiDream-I1_ComfyUI → |
| hidream-i1-dev-BF16.gguf | BF16 | 34.22 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q8_0.gguf | Q8_0 | 18.69 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q6_K.gguf | Q6_K | 14.68 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q5_1.gguf | Q5_1 | 13.52 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q5_K_M.gguf | Q5_K_M | 13.04 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q5_0.gguf | Q5_0 | 12.73 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q5_K_S.gguf | Q5_K_S | 12.49 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q4_K_M.gguf | Q4_K_M | 11.49 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q4_1.gguf | Q4_1 | 11.45 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q4_K_S.gguf | Q4_K_S | 10.90 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q4_0.gguf | Q4_0 | 10.66 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q3_K_M.gguf | Q3_K_M | 8.77 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q3_K_S.gguf | Q3_K_S | 8.22 GB | city96/HiDream-I1-Dev-gguf → |
| hidream-i1-dev-Q2_K.gguf | Q2_K | 6.57 GB | city96/HiDream-I1-Dev-gguf → |
17B sparse MoE DiT. Uses FOUR text encoders (clip_l, clip_g, T5-XXL, Llama-3.1-8B) - heavy offloading. Fast variant (hidream_i1_fast_*) has identical sizes in Comfy-Org repo.