Skip to main content

ComfyUI / Workflow template

Turn a character portrait into a short H3 LoRA video

Start with a character image you already like. This downloadable ComfyUI graph sends it to MiniMax H3 LoRA, adds optional style weights and returns a video to SaveVideo. Use it for moving lookbook covers, original adult-character teasers and short story openings without loading the H3 checkpoint on your computer.

Download workflow JSON

Import the JSON, select your credentials and fill the required inputs before running.

Cinematic storyboard of the same adult male character turning on a rooftop across three frames.
Illustration: a character portrait, a gentle turn and a wider closing frame.
Try the model first
Start with
One starting image, a motion prompt and an optional H3-compatible LoRA URL.
Get
A VIDEO connection saved under the spicyapi/h3-lora filename prefix.
Best for
Character teasers, fashion loops and original adult visual-story openings.

The template contains one hosted generation node. Your computer handles the graph, input image and saved clip; SpicyAPI runs the model. A blank LoRA URL leaves the LoRA stack empty, so you can establish a baseline clip before adding weights. There is no KSampler or local checkpoint loader in this graph.

How the workflow runs

How the workflow runs

  1. 01

    Load starting frame

  2. 02

    Add optional H3 LoRA

  3. 03

    Generate a short clip

  4. 04

    Save the video

A visual guide to the workflow.

NSFW / Character / Style

Creative starting points

Photo-to-video teaser frames
Illustration

Photo-to-video teaser frames

01

Character teasers, fashion loops and original adult visual-story openings.

What to change first

These change the quote. Start with 480p / 5 seconds, then increase one setting after reviewing motion.

Continue to image-to-video
Male creator lookbooks
Illustration

Male creator lookbooks

02

Create a recognizable male character across relaxed portraits, silk-shirt editorials and night-out looks. Keep a few signature traits so the gallery feels like one creator rather than a collection of unrelated faces.

What to change first

Keep the same reference and hair description. Change lighting or wardrobe first, then try a different framing.

Continue to image-to-video
BL & GL story frames
Illustration

BL & GL story frames

03

Build a romantic visual story around two fictional adult characters. Start with a composition that already places both people where you want them, then develop the mood, wardrobe and setting.

What to change first

Use distinct hair, outfits and left/right positions for each person. Inspect both faces before using the frame again.

Continue to image-to-video

Workflow tutorial

Steps to your first result

  1. 01

    Import the graph and connect your account

    Install the SpicyAPI custom-node extension, restart ComfyUI, then drag in the JSON download. Add your key in Settings → SpicyAPI. Update ComfyUI if SaveVideo is missing. The key lives in your settings or server environment, rather than inside the shared template.

  2. 02

    Choose a frame with room for movement

    Select an image in LoadImage. For a face-led teaser, start with a clear face, an uncomplicated pose and space around the shoulders. Use an original fictional adult character or an image you are authorized to upload. Extreme crops leave little room for the movement you ask the model to invent.

  3. 03

    Run a short baseline, then introduce the LoRA

    The SpicyAPI LoRA helper is already connected to the H3 node. Leave its URL blank for the baseline; afterward, supply a direct, publicly accessible H3-compatible weights file and its intended trigger wording. Start with the creator’s suggested strength. A Qwen or FLUX LoRA targets a different base and is not interchangeable with H3 weights.

  4. 04

    Describe motion and choose the clip settings

    Fill the empty prompt with one subject action, a camera instruction and what should stay consistent. The starter uses 480p and 5 seconds. Supported resolution choices are 480p, 540p, 768p and 1080p; duration_seconds accepts 3–15. Keep the seed fixed while changing one setting. Leave last_image_url disconnected unless you have a deliberate ending frame.

  5. 05

    Generate once and save the returned video

    Set a suitable Max Cost Per Run in SpicyAPI settings, then queue the graph once. The hosted node uploads the frame, requests a quote and waits for the task. Its VIDEO output connects directly to SaveVideo. Look in your configured ComfyUI output directory under spicyapi/h3-lora. Keep the task ID if you stop waiting or need to recover the result.

Try this prompt

Creative example. Adapt it to your own fictional adult character and source image.

Keep the fictional adult character, his charcoal coat and the rooftop setting from the starting frame. He slowly turns toward the city lights while his scarf moves gently in the breeze. Static medium-shot camera, subtle movement, consistent face and warm evening light.
Try this direction

Choose a model for your creative range

Choose H3 LoRA when you already have an opening image and want to add a compatible video LoRA. Try the starting frame without a LoRA first, then add your style or character adapter and compare the same motion.

Try the model firstCheck LoRA compatibility

Budget the complete run

One execution submits one H3 video task. Resolution and clip length affect its quote; Max Cost Per Run is a limit for that hosted node, not a daily spending cap. Compare the live price before a longer or higher-resolution run. If you stop ComfyUI after submission, inspect the existing task before queuing again: stopping the local wait does not cancel the accepted task.

See current model pricing

If the result is not what you expected

ComfyUI — If the result is not what you expected

The LoRA file cannot be loaded

Open the exact weights URL without being signed in. A model page URL or gated download is not a direct public file. Check that the weights target H3 and match any trigger words supplied by their creator.

The character moves too much

Return to one modest action and a static camera. Review the baseline without LoRA, then change only strength or prompt. Two simultaneous actions make it harder to tell what caused the drift.

The graph finishes but no MP4 appears

Inspect the VIDEO link into SaveVideo and your configured output directory. A disconnected saver cannot write the hosted result. Use the task ID to locate a result that completed after the local wait ended.

Before you import

Continue with another workflow

Make the first useful run

Download the template or try the same model in the Playground before connecting the full workflow.

Download workflow JSONGet an API key