FIELD GUIDE / 01

Run Hy-Nomas
in ComfyUI.

A practical path from model files to an example workflow, based on the port author’s documentation.

THIRD-PARTY PORT · CHECKED 2026-10-29
Before you begin

This is an early, unofficial integration. Node inputs may change. This site has not run or measured the model itself. The author lists ComfyUI 0.42 as tested, 32 GB system RAM and an NVIDIA GPU; smaller VRAM budgets were simulated.

Another independent port, ComfyUI-HyNomas by Nynxz , uses different files and is not covered by this guide.

Try the model before installing.

Explore Tencent’s official demo ↗ before downloading the model files. The demo runs remotely, so it does not predict performance on your own machine; see local hardware reports.

The port author says a camera path exported from Tencent’s demo can be pasted into the Camera Recording node under { } JSON.

Source: Port camera controls ↗ · checked 2026-10-29

Add the node pack.

In ComfyUI-Manager, search for Hy-Nomas (world model), install it and restart ComfyUI. Or use the Comfy Registry command comfy node install hy-nomas, then restart. Alternatively, clone the repository into your custom nodes directory:

cd ComfyUI/custom_nodes
git clone https://github.com/PedroMarinhoDev/ComfyUI-Hy-Nomas

Source: port quick start .

Put each file in place.

FileDestinationSizeSource
hy_nomas_distil_int8.safetensorsmodels/diffusion_models9.06 GBPort model files
hy_nomas_pixelumm_fp8.safetensorsmodels/vae9.13 GBPort model files
qwen_2.5_vl_7b_fp8_scaled.safetensorsmodels/text_encoders9.38 GBComfy-Org

Model files total approximately 27.6 GB; allow extra space for ComfyUI, cache and output. File sizes use decimal GB. Verified 2026-10-29.

The port author states that the two Hy-Nomas checkpoints are byte-identical to Tencent’s original files and were renamed for ComfyUI. See the port quick start for the original paths. The text encoder is available from Comfy-Org; a BF16 alternative is also available.

hf download PedroMarinhoDev/Hy-Nomas-ComfyUI hy_nomas_distil_int8.safetensors --local-dir ComfyUI/models/diffusion_models
hf download PedroMarinhoDev/Hy-Nomas-ComfyUI hy_nomas_pixelumm_fp8.safetensors --local-dir ComfyUI/models/vae
hf download Comfy-Org/Qwen-Image_ComfyUI split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors --local-dir ComfyUI/models/text_encoders
Looking for the encoder? The last command creates nested text_encoders/split_files/text_encoders/ folders. The port author says ComfyUI can find it there.

Open an example.

Drag Hy-Nomas world (per segment).json from the port’s workflows folder into ComfyUI. Each workflow contains a read-me note.

The author lists the distilled model settings as euler / simple / 8 steps / cfg 1, at 24 fps. The example connects the prompt and text encoder to Segment, KSampler and Extend World, then decodes the resulting world.

This is a summary of the author’s documentation, not an independently tested wiring diagram. Follow the current example workflow for exact connections.

Steer the camera.

In Camera Recording, record a path with W/A/S/D and the arrow keys, or drag the view. Camera Keys accepts a typed timeline such as W 2, W+Left 1.5, idle 0.5.

Paste a path exported from Tencent’s demo into Camera Recording under { } JSON. The port’s Drive node also lets you steer with the keyboard as it generates turns; it is not real time.

Source: Port camera and Drive documentation ↗ · checked 2026-10-29

Set your timing expectations.

These timings were reported by the port author on an RTX 4090, not measured by this site. Results vary by hardware and settings.

For a 20-second world at 1280 × 704, the author reports approximately 17 seconds of sampling and 5 seconds for the cache pass per 5-second segment, plus 91 seconds of decoding.

Port author’s RTX 4090 reports
ExampleTime
20-second world at 1280 × 704, 4 segmentsAbout 3 min 15 s
Start From Image overhead2–6 s
One 1-second Drive turn at 832 × 480, general reportAbout 3.5–4 s
One 1-second Drive turn at 832 × 480, sample clipAbout 4.3 s

Longer worlds also increase history stored in system RAM; see hardware reports and the RAM section.

Source: Port speed and memory notes ↗ · checked 2026-10-29

The separate sample-clip timing appears in the samples section ↗.

Check the basics.

The text encoder does not appear

Confirm the download completed and check the nested folder described above. Restart ComfyUI and consult the port’s current quick start.

I see an out-of-memory error

Check system RAM as well as VRAM. The author reports that history grows with longer worlds. Consult the hardware page and the port’s documentation for current settings.

A workflow or node fails

Confirm your ComfyUI and node pack versions, then consult the port’s issues. Include the full error when reporting a problem; a black screen alone does not establish a cause.