Datasheet for local AI50 models99 GPUsData read 2026-10-01
No adsNo tracking
Check my GPU →

Free ComfyUI workflows, tested on a 16 GB card

12 ready-to-run workflows for the models people ask about most — FLUX.1, FLUX.2 klein, Z-Image, Qwen-Image 2.1, Krea 2, Ming-Image and Wan 2.2. Each one ran on my RTX 5060 Ti 16 GB, and each opens with a note that lists every file, where it goes, and how fast it was.

Tested on RTX 5060 Ti 16 GB32 GB RAMComfyUI 0.38.1Free

Download all 12 workflows (.zip, 32 KB)

Same prompt, same seed, 9 models

What each workflow makes out of the box, straight from my test runs (seed 42, no cherry-picking). Ming-Image gets a poster prompt, because that is what it is made for.

FLUX.1 [dev] sample image: a cozy coffee shop on a rainy evening, warm light through the window, a barista pouring latte art, film photo, shallow depth of field
FLUX.1 [dev] FP8
36.4 s / image
FLUX.1 [dev] sample image: a cozy coffee shop on a rainy evening, warm light through the window, a barista pouring latte art, film photo, shallow depth of field
FLUX.1 [dev] GGUF Q8_0
40.6 s / image
FLUX.1 [schnell] sample image: a cozy coffee shop on a rainy evening, warm light through the window, a barista pouring latte art, film photo, shallow depth of field
FLUX.1 [schnell] GGUF Q8_0
9.4 s / image
FLUX.2 klein 4B sample image: a cozy coffee shop on a rainy evening, warm light through the window, a barista pouring latte art, film photo, shallow depth of field
FLUX.2 klein 4B FP8
2.9 s / image
FLUX.2 klein 9B sample image: a cozy coffee shop on a rainy evening, warm light through the window, a barista pouring latte art, film photo, shallow depth of field
FLUX.2 klein 9B GGUF Q8_0
8.5 s / image
Z-Image Turbo sample image: a cozy coffee shop on a rainy evening, warm light through the window, a barista pouring latte art, film photo, shallow depth of field
Z-Image Turbo 16-bit
11.0 s / image
Qwen-Image 2.1 sample image: a cozy coffee shop on a rainy evening, warm light through the window, a barista pouring latte art, film photo, shallow depth of field
Qwen-Image 2.1 INT8
17.2 s / image
Krea 2 Turbo sample image: a cozy coffee shop on a rainy evening, warm light through the window, a barista pouring latte art, film photo, shallow depth of field
Krea 2 Turbo INT8
10.9 s / image
Ming-Image 0.1 Design sample image: minimalist poster for a coffee brand called "Summit Brew", mountain silhouette, bold sans-serif title, cream background, flat design
Ming-Image 0.1 Design INT8
7.4 s / image

Video: one frame from each test clip

Wan 2.2 TI2V 5B video frame: a red fox walking through fresh snow in a pine forest, slow camera pan, soft morning light, cinematic
Wan 2.2 TI2V 5B text to video
9 min for a 121-frame 1280×704 clip
Wan 2.2 T2V 14B video frame: a red fox walking through fresh snow in a pine forest, slow camera pan, soft morning light, cinematic
Wan 2.2 T2V 14B FP8
19 min for an 81-frame 832×480 clip
Wan 2.2 T2V 14B video frame: a red fox walking through fresh snow in a pine forest, slow camera pan, soft morning light, cinematic
Wan 2.2 T2V 14B GGUF Q5_K_M
26 min for an 81-frame 832×480 clip

How to use them

  1. Download a .json file below, or the zip with all of them.
  2. Drag it into the ComfyUI window. The workflow opens with a READ ME note on the left.
  3. The note lists each file with a download link and the folder it goes in (ComfyUI/models/diffusion_models, text_encoders or vae). Download them, then press R in ComfyUI to refresh the file lists.
  4. GGUF workflows need the ComfyUI-GGUF custom node: open Manager → Custom Nodes Manager, search “ComfyUI-GGUF”, install, restart.
  5. Qwen-Image 2.1, Krea 2 and Ming-Image need ComfyUI 0.38 or newer.

Not sure a model fits your card? Check your GPU first. These workflows use the file that runs best on a 16 GB card; on a smaller card, take the smaller GGUF the check page suggests and pick it in the loader node.

The workflows

Model and fileMeasured on RTX 5060 Ti 16 GBPeak VRAMDownload size
FLUX.1 [dev]
FP8
4 filesflux1-dev-fp8-e4m3fn.safetensors
t5xxl_fp8_e4m3fn.safetensors
clip_l.safetensors
ae.safetensors
36.4 s per 1024×1024 image14.8 GB17.4 GBDownload .json
FLUX.1 [dev]
GGUF Q8_0
4 filesflux1-dev-Q8_0.gguf
t5xxl_fp8_e4m3fn.safetensors
clip_l.safetensors
ae.safetensors
40.6 s per 1024×1024 image15.1 GB18.2 GBDownload .json
FLUX.1 [schnell]
GGUF Q8_0
4 filesflux1-schnell-Q8_0.gguf
t5xxl_fp8_e4m3fn.safetensors
clip_l.safetensors
ae.safetensors
9.4 s per 1024×1024 image15.1 GB18.2 GBDownload .json
FLUX.2 klein 4B
FP8
3 filesflux-2-klein-4b-fp8.safetensors
qwen_3_4b_fp8_mixed.safetensors
flux2-vae.safetensors
2.9 s per 1024×1024 image12.2 GB10.0 GBDownload .json
FLUX.2 klein 9B
GGUF Q8_0
3 filesflux-2-klein-9b-Q8_0.gguf
qwen_3_8b_fp8mixed.safetensors
flux2-vae.safetensors
8.5 s per 1024×1024 image13.5 GB19.0 GBDownload .json
Z-Image Turbo
16-bit
3 filesz_image_turbo_bf16.safetensors
qwen_3_4b_fp8_mixed.safetensors
ae.safetensors
11.0 s per 1024×1024 image14.6 GB18.3 GBDownload .json
Qwen-Image 2.1
INT8
3 filesqwen_image_2.1_int8_convrot.safetensors
qwen3vl_8b_int8_convrot.safetensors
qwen_image_2.1_vae_bf16.safetensors
17.2 s per 1024×1024 image14.3 GB17.3 GBDownload .json
Krea 2 Turbo
INT8
3 fileskrea2_turbo_int8_convrot.safetensors
qwen3vl_4b_fp8_scaled.safetensors
qwen_image_vae.safetensors
10.9 s per 1024×1024 image15.3 GB19.0 GBDownload .json
Ming-Image 0.1 Design
INT8
3 filesming_image_0.1_design_int8_convrot.safetensors
ming_image_0.1_ling_mini_2.0_int8_convrot.safetensors
ming_image_vae_bf16.safetensors
7.4 s per 1024×1024 image13.6 GB25.9 GBDownload .json
Wan 2.2 TI2V 5B
text to video
3 fileswan2.2_ti2v_5B_fp16.safetensors
umt5_xxl_fp8_e4m3fn_scaled.safetensors
wan2.2_vae.safetensors
9 min for a 121-frame 1280×704 clip11.0 GB18.2 GBDownload .json
Wan 2.2 T2V 14B
FP8
4 fileswan2.2_t2v_high_noise_14B_fp8_scaled.safetensors
wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors
umt5_xxl_fp8_e4m3fn_scaled.safetensors
wan_2.1_vae.safetensors
19 min for an 81-frame 832×480 clip13.3 GB35.6 GBDownload .json
Wan 2.2 T2V 14B
GGUF Q5_K_M
4 filesWan2.2-T2V-A14B-HighNoise-Q5_K_M.gguf
Wan2.2-T2V-A14B-LowNoise-Q5_K_M.gguf
umt5_xxl_fp8_e4m3fn_scaled.safetensors
wan_2.1_vae.safetensors
26 min for an 81-frame 832×480 clip14.0 GB28.6 GBDownload .json

Times are from my own runs with default ComfyUI settings: the average of 2–3 runs after a warm-up, sampling plus VAE decode. Video times include loading both Wan 2.2 14B experts. Peak VRAM is the memory in use on the whole card. Every number: measured results and the RTX 5060 Ti 16 GB guide.

Licence

The workflow files are free to use, change and share, including commercially. A link back is appreciated. The models have their own licences — FLUX.1 [dev] and FLUX.2 klein 9B, for example, are non-commercial; check each model page.