Choose the right label

<Subject 1> names reusable visible content, such as a person, outfit, scene, or style. One upload can define several subjects.
<Picture 1> names an image used as a first frame, keyframe, final frame, or composition anchor. For appearance alone, define a subject from that picture.
<Video 1> names a source to edit, continue, or borrow camera movement and timing from. Name visible people or objects from it as subjects.
<Audio 1> names audio to copy or use for timbre, rhythm, or sound style. Video and audio numbers are independent; an included video soundtrack takes the first audio label in this form.

State how you use it

Open a structured summary with the roles that apply: [reference generation], [keyframe completion], [video editing], [video continuation], [audio reuse], or [audio reference]. Join multiple roles with + .

Use [reference generation] when a video guides camera movement, cuts, or rhythm. Use [video editing] when changing the source clip itself, or [video continuation] when extending it.

For visible references, describe fully_preserved, partially_preserved, attribute_transfer, or weak_reference. For sound, use fully_copy, partially_copy, reference, or weak_reference. Use a copy term only when the audio signal itself is reused.

Build a scene

Use [Shot 1] for the opening and [Shot 2] At 00:03.000 for a cut. Describe the visible change and current sound in each shot. Give a speaker a stable ID such as (S1); put spoken words inside <d>[English] ...</d>.

This example excludes the video's soundtrack, so the uploaded piano uses <Audio 1>. With video audio included, the piano would use <Audio 2>.

subject_definitions:
<Subject 1> is the cyclist in <Picture 1>, with a blue jacket and silver helmet.
<Video 1> supplies the handheld tracking movement.
<Audio 1> supplies a steady piano rhythm.

summary:
[reference generation + audio reference] <Subject 1> rides through a wet city street.

retention_analysis:
<Subject 1>: fully_preserved - keep the jacket, helmet, and face.
<Video 1>: weak_reference - follow its camera motion, with a new setting.
<Audio 1>: reference - match the rhythm without copying the recording.

detailed_description:
[Shot 1] A low tracking shot follows <Subject 1> across reflective pavement under soft dusk light. The piano rhythm builds as tires hiss on the road.

overall_soundscape: Tires and distant traffic.
non_diegetic_music: Soft piano with a steady pulse.

For a longer reference-heavy prompt, use these six sections. A short prompt can simply describe the intended shot and refer to the same labels.

PixForge is a free AI image and video generator. Create with Illustrious and Wan 2.2 using Image-to-Video, Img2Img, inpainting, LoRAs, and more.

Community

Notifications

Create

Anonymous generator feedback Share favourite models, LoRAs, generator issues, and what you want added next.

Give Feedback

Create a free account for faster generations and daily credits.

Earn daily rewards and bonus credits
Fast queue access and generation history
Save slots and LoRA presets across devices

First time? Generate something in three quick steps.

Describe what you want in the Prompt box. Specific details help.
Pick an Aspect Ratio and Model that match your idea.
Press Generate Image. You can tweak advanced options later.

Try a ready-made prompt:

Save Slots (click to open) (0) Store prompt presets for quick reuse (advanced)

You must be logged in to create a save slot!

Save Slots Help:

Save-Slots are a way of saving and loading the current settings, very good for saving templates to generate the best images for later use, this requires you to be logged in to use.

If you'd like to save your settings without logging in then you can use the "Export/Import .AIDATA" buttons beneath this text to save your settings to a file on your computer/phone!

Export & Import Settings (click to open) Backup or share generator configurations

Prompt & Negative Prompt

Prompt:

Type here what you want in the image. You can use "," to separate different things you want in the image. For example: "A cute cat, with a hat, in a forest"

To adjust how much focus or Strength specific word(s) have you can use the following syntax:
(word), ((word)), ((word1, word2))
(word:1.25), ((word:1.25)), ((word1, word2:1.25))
(word:0.75), ((word:0.75)), ((word1, word2:0.75))

These are just examples of setting the strength(s) of the word(s) in the prompt, you can use any number between 0.1 and 3.0 to adjust the strength of the word(s) in the prompt.

Negative Prompt:

Type here what you don't want in the image. You can use "," to separate different things you don't want in the image. For example: "ugly, dogs, no hat, no forest"

To adjust how much focus or Strength specific word(s) have you can use the following syntax:
(word), ((word)), ((word1, word2))
(word:1.25), ((word:1.25)), ((word1, word2:1.25))
(word:0.75), ((word:0.75)), ((word1, word2:0.75))

These are just examples of setting the strength(s) of the word(s) in the prompt, you can use any number between 0.1 and 3.0 to adjust the strength of the word(s) in the prompt.

Aspect Ratio & Model

Aspect Ratio:

Select the aspect ratio you want the image to be in, this will affect the dimensions of the image.

Model:

Select the model you want to use to generate the image.
Certain models have different architectures and will need different prompts to generate the desired image, this can be denoted by the keyword in brackets at the start of the model.
For example: "(Illustrious) AutismMix" is an Illustrious architecture model.

LoRA Add-ons (click to open) (0/0) Attach style, character, or effect boosters

Choose up to 3 Minimax H3 LoRAs. Only LoRAs made for H3 appear here.

Lora Search:

Use this field to search for a specific Lora by name, use the buttons below to search by category.

Lora Safety Ratings:

Use these checkboxes to filter Loras by their safety ratings, with explicit loras being hidden behind the "suggestive" and "nsfw" ratings.

SFW Suggestive NSFW

Lora Sorting:

Use these buttons to sort the Loras. Trending shows LoRAs with recent usage activity.

Mix safety: SFW Suggestive NSFW

Open Mixes to load presets for the selected model.

Enable Image2Image:

Use this checkbox to enable the Image2Image feature, which allows you to upload an image to be used by the AI Image Generator

Enable Inpainting:

Use this checkbox to enable the Inpainting feature, which allows you to upload an image to be used by the AI Image Generator
Inpainting lets you paint over an image to change the content that's painted over

Source editor

Paint the area to regenerate

The translucent cyan overlay keeps the source visible while you edit.

No image selected

Source image

Generation strength 70% Higher values replace more of the painted area.

Drag to paint the mask, or select Fill area and click inside an enclosed outline. Right-click and drag always erases with the brush, even while Fill area is selected.

Guidance & Seed (click to open) Fine-tune CFG strength and seeding

CFG:

Use this slider to adjust the CFG guidance value
Here are the recommended values for the different model architectures:
ILLUSTRIOUS: 2-3.5 Anima: 1-1.5
Wan: 1 (locked)

/0.3

Seed: (-1 = random)

Use this input to set the seed for the AI, -1 is random
The seed is used to generate the image, so the same seed will always generate the same image aslong as the other settings are the same

IMG Compression: 10

IMG Compression alters how much the end result moves. Higher values provide more movement with less quality, while lower values provide better quality with less movement.

5 40

💳 Total Cost: 0 Credits

Optional extras use credits you earn or buy. The total above updates for your selected model and options. Fast queue gives priority in the queue. Compare credit packs here.

🏷️ Remove Watermark — Credits

📈 Upscale Image — Credits

🎭 Remove Background — Credits

⚡ Fast Queue — Credits

LoRAs: 0

CFG: 6

Seed: -

Generate Image:

Use this button to generate the image using the AI Image Generator
The image will be generated based on the settings you have selected

Output

Choose a Model

Give each reference a job

Choose the right label

State how you use it

Build a scene

Build a stronger video prompt

Write for the whole clip

Define the picture

Direct motion and camera

Plan dialogue and sound

Text-to-video

Image-to-video

Reliable strengths

Common trouble spots

Useful production language

Example structure

Loading...

Notifications

Create

First time? Generate something in three quick steps.

Minimax H3 references

Pictures

Audio

Video

Paint the area to regenerate

Community Guidelines

Prohibited Content

Age-Up Guidelines

Moderation Process

Quality Standards

Consequences

Loading...

Create

First time? Generate something in three quick steps.

Create Mix

Minimax H3 references

Pictures

Audio

Video

Community Guidelines

Prohibited Content

Age-Up Guidelines

Moderation Process

Quality Standards

Consequences