✅ Complete Beginner’s Guide: How to Set Up and Use the LTX Video Workflow

(Made with help from Grok – blame El00n if the PC explodes)

@LoRa_Daddy – Please feel free to delete or edit if anything is incorrect.

### What This Workflow Does

This tool inside ComfyUI lets you create short cinematic videos. You give it a photo of a person and a few words. It automatically adds music, dialogue, camera movements, and special effects — like having a real film director inside your computer. Perfect for funny comedy videos or over-the-top action-comedy scenes (including Spooderman spoofs).

### Step 1: Install ComfyUI (The Main Program)

1. Go to this website: https://github.com/comfyanonymous/ComfyUI/releases

2. Download the latest portable version for Windows.

3. Extract the zip file to a new folder (example: C:\ComfyUI).

4. Double-click run_nvidia_gpu.bat (or run_cpu.bat if you don’t have a good graphics card).

A black window will open. Wait until it finishes.

### Step 2: Install ComfyUI Manager

1. Open the folder: C:\ComfyUI\ComfyUI\custom_nodes

2. Download this file: https://github.com/ltdrdata/ComfyUI-Manager/archive/refs/heads/main.zip

3. Extract it and rename the folder to ComfyUI-Manager.

4. Close and reopen ComfyUI.

5. You will now see a Manager button at the bottom right.

### Step 3: Install the Required Custom Nodes

- Click the Manager button.

- Search and install: KJNodes, VideoHelperSuite, and ComfyUI-LTXVideo.

For the two special LTX nodes:

1. Open a terminal window inside the custom_nodes folder.

2. Copy and paste these two commands one by one:

```

git clone https://github.com/seanhan19911990-source/LTX2EasyPrompt-LD.git

git clone https://github.com/seanhan19911990-source/LTX2-Master-Loader.git

```

3. Restart ComfyUI.

### Step 4: Download and Place the Required Models

Most models are downloaded automatically by the workflow, so you don’t need to hunt for them manually. However, placing these files manually gives the best results:

- Main Video Model → C:\ComfyUI\ComfyUI\models\diffusion_models\

Download: https://huggingface.co/Kijai/LTX2.3_comfy/resolve/main/diffusion_models/ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled_v2.safetensors

- Preview VAE (Note: This is not the main VAE used by the workflow) → C:\ComfyUI\ComfyUI\models\vae\

Download: https://huggingface.co/Kijai/LTX2.3_comfy/resolve/main/vae/taeltx2_3.safetensors

- Spatial & Temporal Upscalers → C:\ComfyUI\ComfyUI\models\latent_upscale_models\

- Spatial Upscaler: https://huggingface.co/Lightricks/LTX-2.3/resolve/main/ltx-2.3-spatial-upscaler-x2-1.1.safetensors

- Bonus (recommended by the author): Also download the Temporal Upscaler and place it in the same latent_upscale_models folder. This helps the workflow output double FPS videos.

Restart ComfyUI after placing the files.

### Step 5: Install RTX Super Resolution (For Sharper Videos)

1. Go to your ComfyUI folder and open the python_embeded folder.

2. Right-click inside it and select “Open Terminal here”.

3. Copy and paste this exact command, then press Enter:

```

.\python.exe -m pip install -U --no-build-isolation nvidia-vfx --index-url https://pypi.nvidia.com

```

4. Restart ComfyUI.

### Step 6: Set Up the AI Models (So They Never Download Again)

The workflow uses two different AI models that work together:

- prithivMLmods/Qwen2.5-VL-7B-Abliterated-Caption-it — This is the image reading model (shown as LTX-2 Vision Describe By LoRa-Daddy). It reads your reference photo and understands the character, clothing, pose, and context.

- huihui-ai/Huihui-Qwen3.5-9B-abliterated — This is the prompt generation model (shown as LTX-2.3 Easy Prompt Qwen By LoRa-Daddy). It takes your short text prompt + the image information and creates a detailed cinematic prompt and dialogue.

To stop them from re-downloading every time:

1. The paths will look similar to these (the long number at the end may be different on your PC):

- Prompt model: C:\Users\USER\.cache\huggingface\hub\models--huihui-ai--Huihui-Qwen3.5-9B-abliterated\snapshots\05b9e7c9b978ba29bdb8f50a49c30e4b91183339

2. Make sure each snapshot folder contains the four large model.safetensors-0000x-of-00004.safetensors files.

3. In the LTX-2.3 Easy Prompt Qwen By LoRa-Daddy node, set offline_mode to True.

If models are missing the first time, set it to False temporarily, run once, then set it back to True.

### Step 7: Load Your Workflow

1. In ComfyUI, click Load and select your workflow JSON file.

2. If any nodes are red, click Manager → Install Missing Custom Nodes.

3. Restart ComfyUI.

### Step 8: Custom Audio – Use Your Own Song or Let It Generate

There is no switch named “Custom Audio”. Look for the group called:

> Audio - Trimmed to length of video

Inside this group, find the small round switch button (labelled switch).

- To use your own audio file: Set the switch to True and load it in the Load Audio node.

- To let the AI create its own music and dialogue (recommended): Set the switch to False.

You will also see the MelBandRoFormer Model Loader (model: MelBandRoFormer_fp32.safetensors).

### Step 9: Best Settings for Spooderman Video

In the node LTX-2.3 Easy Prompt Qwen By LoRa-Daddy:

- Use image information? → True

- Let the LLM create dialogue? → True

- Creativity → 0.85

- Shot angle → High angle — vulnerable

- Camera movement → Slow push in

- Music Genre → Comedy or Pop

- Emotional state → Joyful or Playful

- Reference Image Strength → 0.92

### Step 10: Make Your First Spooderman Video

1. Load your Spooderman reference image in the LoadImage node.

2. Paste this prompt:

```

over-the-top hilarious Spooderman spoof action scene, ridiculous parody of Spider-Man, clumsy hero named Spooderman wearing a cheap red-and-blue suit with visible zippers and loose threads, oversized web shooters that spray silly string instead of webs, terrible aim, constantly tripping over his own feet, accidentally swinging into lampposts and trash cans, comically failing at heroic poses, dramatic slow-motion falls, exaggerated facial expressions, goofy sound effects, chaotic city street at night, cars honking, pedestrians laughing and filming him, buildings covered in silly string, villain is just a confused robber in a cheap costume, pure slapstick comedy, silly action parody, very funny, light-hearted, cartoonish chaos

```

3. Use the Spooderman settings from Step 9.

4. Click Queue Prompt.

### Step 11: Quick Troubleshooting

- Red nodes → Install missing custom nodes again.

- No sound → Set the switch in Audio - Trimmed to length of video to False.

- Model not loading → Check offline_mode and restart.

- Blurry video → Enable RTX upscaler + use both Spatial and Temporal upscalers in latent_upscale_models.

You now have everything you need! The workflow itself doesn’t change much anymore — only capabilities and presets may update in the future.

Take your time and enjoy making funny Spooderman videos!

That kinda-scottish accent on the preview vid makes it so funny 😂

the face isn't the same as i uploaded. Spooderman lost his mask and now all his friends bully him at school! Not fair! I want justice for Spooderman! Or a way to control denoise strength, steps and cfg. and yes, I have a backup keyboard. for times when I need to fight for my heroes! for Spooderman!

could you help me with this problem?

ValueError: The checkpoint you are trying to load has model type qwen3_5 but Transformers does not recognize this architecture. This could be because of an issue with the checkpoint, or because your version of Transformers is out of date.

You can update Transformers with the command pip install --upgrade transformers. If this does not work, and the checkpoint is very new, then there may not be a release version that supports this model yet. In this case, you can get the most up-to-date code by installing Transformers from source with the command pip install git+https://github.com/huggingface/transformers.git

File "C:\Users\artur\AppData\Local\Programs\ComfyUI\resources\ComfyUI\execution.py", line 524, in execute

output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\artur\AppData\Local\Programs\ComfyUI\resources\ComfyUI\execution.py", line 333, in get_output_data

return_values = await asyncmap_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\artur\AppData\Local\Programs\ComfyUI\resources\ComfyUI\execution.py", line 307, in asyncmap_node_over_list

await process_inputs(input_dict, i)

File "C:\Users\artur\AppData\Local\Programs\ComfyUI\resources\ComfyUI\execution.py", line 295, in process_inputs

result = f(**inputs)

^^^^^^^^^^^

File "C:\Users\artur\Documents\ComfyUI\custom_nodes\LTX2EasyPrompt-LD-Pre-Extra-feature-Main\LTX2EasyPromptQwen.py", line 2747, in generate

self.load_model(offline_mode=offline_mode, local_path=local_path, model_id=selected_model)

File "C:\Users\artur\Documents\ComfyUI\custom_nodes\LTX2EasyPrompt-LD-Pre-Extra-feature-Main\LTX2EasyPromptQwen.py", line 2372, in load_model

self.model = AutoModelForCausalLM.from_pretrained(

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\artur\Documents\ComfyUI\.venv\Lib\site-packages\transformers\models\auto\auto_factory.py", line 549, in from_pretrained

config, kwargs = AutoConfig.from_pretrained(

^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "C:\Users\artur\Documents\ComfyUI\.venv\Lib\site-packages\transformers\models\auto\configuration_auto.py", line 1362, in from_pretrained

raise ValueError(

When y'all download this, make sure you delete the pycache files. just best security practice

how to disable music genre? if i generate txt2video+audio, why i need to load mp3 and image?

Can GGUF be used for EasyPrompt offline?

Awesome work. I’m a huge fan. However with this version, after it produces the prompt, it ends and doesn’t try to create the video.

Hi, Sean! I still can't make it work correctly :(

I tried different "width", "height" values and also "frame count" more than 300, as you advised me on reddit(my name is Constantine).

Transformers version also is latest (5.3.0).

Can you share arguments for running Comfyui and also list of pip packages with versions?

Costom audio + Use audio for the llm (On or off)

Is not working.. Can you check this for me?