Hunyuan / IMAGE
HunyuanImage 2.1 VRAM requirements
Explore memory tradeoffs with sourced model data and transparent estimates.
FP8 estimate
19.9 GiB
at default settings
FP16 estimate
38.0 GiB
at default settings
Suggested GPU
1 × RTX 4000 Ada Generation 20 GiB
Model files & facts
| Media | image |
|---|---|
| Default resolution | 2048 × 2048 |
| Default frames | 1 |
| Temporal compression | Not applicable |
| Notes | Models the base denoiser pipeline. The optional refiner is a separate processing stage. |
| Component | Weight file (source) | Size |
|---|---|---|
| Backbone FP16 | split_files/diffusion_models/hunyuanimage2.1_bf16.safetensors | 32.46 GiB |
| Backbone FP8 | split_files/diffusion_models/hunyuanimage2.1_fp8_e4m3fn.safetensors | 16.23 GiB |
| Qwen2.5-VL 7B FP16 | split_files/text_encoders/qwen_2.5_vl_7b.safetensors | 15.45 GiB |
| Qwen2.5-VL 7B FP8 | split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors | 8.74 GiB |
| ByT5 Small FP16 | split_files/text_encoders/byt5_small_glyphxl_fp16.safetensors | 0.41 GiB |
| VAE | split_files/vae/hunyuan_image_2.1_vae_fp16.safetensors | 0.76 GiB |
Precision & offloading comparison
Source default resolution and frame count · batch 1 · FP16 text encoders · safety margin included.
| Backbone | Text encoders | Total |
|---|---|---|
| FP16 | GPU | 38.03 GiB |
| FP16 | CPU | 38.03 GiB |
| FP8 | GPU | 19.85 GiB |
| FP8 | CPU | 19.85 GiB |
ComfyUI file placement
Example component locations; install matching loader nodes. Brace patterns in source filenames denote all shards. Preserve the source layout when a loader requires it.
models/diffusion_models/split_files/diffusion_models/hunyuanimage2.1_bf16.safetensors
models/diffusion_models/split_files/diffusion_models/hunyuanimage2.1_fp8_e4m3fn.safetensors
models/text_encoders/split_files/text_encoders/qwen_2.5_vl_7b.safetensors
models/text_encoders/split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors
models/text_encoders/split_files/text_encoders/byt5_small_glyphxl_fp16.safetensors
models/vae/split_files/vae/hunyuan_image_2.1_vae_fp16.safetensorsRecommended GPUs by memory fit
FP8 backbone at source default resolution, FP16 text encoders on GPU. Smallest nominal memory first; green results only. Check backend and precision support separately.
1 × RTX 4000 Ada Generation 20 GiB
Fits with safety headroom
1 × AMD RX 7900 XT (20 GiB)
Fits with safety headroom
1 × RTX 3090 24 GiB
Fits with safety headroom
Sources & verification
Exact file sources are linked in the component table above. ComfyUI memory planner constant
Last verified: 2026-11-20
Try your configuration
Your workload
Custom diffusion model parameters
Uses the ComfyUI activation and decode profile for HunyuanImage 2.1. Leave fields empty to use catalog files. Placeholder counts approximate half-precision file sizes.
Your hardware
Output dimensions
Images or videos per batch. Serving concurrency and throughput require separate measurements.
Resolution and peak VRAM
Sampling and timing
One step is one scheduler iteration. Enter timings from the same device, batch, resolution and guidance settings. Fixed overhead includes encoding, decoding and other work outside sampling.
Advanced
Performance and memory results
Memory estimate
19.9 GiB needed
The estimated workload fits with the safety margin included.
RTX 4090 24 GiB
Vertical line = usable hardware memory
- Diffusion model (FP8)16.23 GiB
- Sampling activations0.70 GiB
- ComfyUI inference reserve0.80 GiB
- Safety margin2.13 GiB
Per-stage peak
ComfyUI moves idle models to system RAM between stages; the largest stage sets the requirement. Totals include the safety margin.
- Text encode18.65 GiB
Text encoders (FP16) 15.85 + ComfyUI inference reserve 0.80
Subtotal 16.65 + safety margin (12%) 2.00 ≈ 18.65 GiB
- SamplingPeak19.85 GiB
Diffusion model (FP8) 16.23 + Sampling activations 0.70 + ComfyUI inference reserve 0.80
Subtotal 17.73 + safety margin (12%) 2.13 ≈ 19.85 GiB
- VAE decode7.87 GiB
VAE 0.76 + VAE decode activations 5.47 + ComfyUI inference reserve 0.80
Subtotal 7.02 + safety margin (12%) 0.84 ≈ 7.87 GiB
Generation performance
- Denoising steps
- 50
- Conditioning branches per step
- 1
- Latent patch tokens per sample
- 4,096
- Estimated backbone work per step
- 157.184 TFLOP
- Step time at single-device BF16 peak
- 951.475 ms
DiT work estimates cover linear layers and visual self-attention. Enter measured step time and fixed overhead under Sampling and timing.
Compare GPU capacity
Frequently asked questions
How much VRAM does HunyuanImage 2.1 need?
Our default estimate at the source default resolution, batch 1, FP16 text encoders on GPU and FP8 backbone is 19.9 GiB including the safety margin. These are estimates, not measured guarantees.
Can I offload text encoders?
Yes. The CPU option removes text encoder weights from the GPU estimate, trading GPU memory for system RAM and transfer time. Each component still needs to fit on its assigned device.
Does a green result guarantee compatibility?
No. Green means the estimated memory including safety headroom fits. Backend support, precision support, kernels and runtime allocation can differ. Confirm compatibility and measure your actual workflow.
Where do these estimates come from?
Architecture and file sizes are traced to model sources. The precision table identifies measured GGUF rates, block storage and nominal bit widths. The source data for HunyuanImage 2.1 was last verified on 2026-11-20. Memory estimates use GiB.