Hardware requirements by model family: use only what official sources state

Purpose and scope

Minimum-VRAM claims are often unsourced. This guide lists only what the official ComfyUI documentation and publisher pages state for SD1.5, SDXL, FLUX.1, Qwen-Image, Wan 2.2 and LTX-Video, with their conditions, and says where no figure exists.

Source-supported context

ComfyUI. The system-requirements page lists OS, GPU vendors, Python and PyTorch but no VRAM, RAM or disk figure (Source 1).

SD1.5 and SDXL. Neither card states a memory requirement; the only GB figure on the SD1.5 card, "A100 PCIe 40GB", is the training hardware in its environmental-impact section. The SD1.5 card says v1-5-pruned-emaonly.safetensors "uses less VRAM - suitable for inference" and v1-5-pruned.safetensors "uses more VRAM" (Source 9); an SDXL Diffusers example suggests pipe.enable_model_cpu_offload if VRAM-limited (Source 8).

FLUX.1. The FLUX.1 [dev] card states no requirement; a Diffusers comment says CPU offloading will "save some VRAM" (Source 6). The ComfyUI tutorial says the full version needs more VRAM and several files, the FP8 checkpoint one file at reduced quality, and the FP16 T5XXL encoder is recommended when VRAM is greater than 32GB, otherwise the FP8 encoder (Source 4). That 32GB concerns the encoder choice only.

Qwen-Image. The card gives no figure (Source 7). The ComfyUI tutorial's "VRAM Usage Reference" lists an RTX4090D 24GB test environment with 86% VRAM usage for fp8_e4m3fn, the same with a lightx2v 8-step LoRA, and Distilled fp8_e4m3fn (Source 3). That is a reference measurement, not a minimum.

Wan 2.2. The official repository says the A14B commands "can run on a GPU with at least 80GB VRAM" and the TI2V-5B command "can run on a GPU with at least 24GB VRAM (e.g, RTX 4090 GPU)" with --offload_model True, --convert_model_dtype, --t5_cpu (Source 5). These are for its own scripts. The README also has a "Computational Efficiency on Different GPUs" section that reports peak GPU memory, but as an image; this guide did not read its values. The ComfyUI tutorial says the 5B version "should fit well on 8GB vram with the ComfyUI native offloading" and gives no 14B figure (Source 2).

LTX-Video. The repository's model table gives no VRAM requirement, only relative wording such as "requires more VRAM" or "less VRAM usage" (Source 10). Its only figure is a May 2025 news item saying the LoRA version of the 13B distilled model "Requires only 1GB of VRAM" (the LoRA, to be used with the full 13B model, not a whole-pipeline requirement). An 8GB RTX 4060 result there belongs to a community project.

Before you start

1. Name the exact variant, precision and file layout.

2. Fix resolution, frames, batch size, offloading.

3. Note GPU, driver, ComfyUI version and launch flags.

Steps and checks

The sequence below is an editorial procedure based on the cited context, not a diagnosis already confirmed for your environment.

Step 1. Copy each official statement with its conditions, e.g. "with the ComfyUI native offloading".

Step 2. Classify each figure: requirement, component recommendation or reference measurement. Write "no official figure" where true.

Step 3. Match the runtime: the Wan repository's numbers apply to its scripts, not ComfyUI offloading.

Step 4. Compare variants only on stated attributes; no source gives a conversion rule.

Step 5. Run your workflow once at the intended resolution and record files, resolution, frames, flags, peak GPU memory from your own monitor, and the outcome (completed, error text or out of memory).

Step 6. Change one load factor at a time and re-record, per the low-memory checklist below, labelled as your measurement.

Completion check

Each model has official wording with link (or "no official figure") and your own dated run record, kept separate.

Limitations and cautions

Sources may change. Quoted figures are from the pages as read on 2026-10-06. A missing figure does not show a model runs on your hardware. This site ran nothing.

Original sources

• ComfyUI system requirements — checked 2026-10-06.
• ComfyUI Wan2.2 tutorial — checked 2026-10-06.
• ComfyUI Qwen-Image tutorial — checked 2026-10-06.
• ComfyUI FLUX.1 tutorial — checked 2026-10-06.
• Wan2.2 official repository — checked 2026-10-06.
• FLUX.1 [dev] model card — checked 2026-10-06.
• Qwen-Image model card — checked 2026-10-06.
• SDXL base 1.0 model card — checked 2026-10-06.
• Stable Diffusion v1-5 model card — checked 2026-10-06.
• LTX-Video official repository — checked 2026-10-06.

Prepared 2026-10-06 from the sources above. No GPU run, quality measurement or runtime guarantee is supplied.

Related troubleshooting and guides

• Tune a low-memory workflow
• Qwen-Image 2.1 low memory
• Wan 2.2 5B versus 14B
• FLUX.1 files and loaders