08 — AI Image Generation
AI Image Generation is available under AI Generate → Image. Generated images can be downloaded directly or added to your timeline as a video segment (still frame).
A Replicate API key is required for all models. Some models additionally require an OpenAI API key.
Available Models
Google Models
| Model | Best For |
|---|---|
| Nano Banana Pro | High quality; supports up to 14 reference images |
| Nano Banana 2 | Faster; supports up to 14 reference images |
Input: aspect ratio selector. Reference images are supported.
Flux (Black Forest Labs)
| Model | Notes |
|---|---|
| Flux 2 Max | Highest quality Flux model |
| Flux 2 Pro | Professional; supports up to 8 reference images |
| Flux 2 Flex | Creative flexibility; guidance range 1.5–10 |
| Flux Dev | Open-weight development model |
| Flux Schnell | Fastest Flux model; good for iteration |
Seedream
| Model | Notes |
|---|---|
| Seedream 4.5 | 2K or 4K output quality selector |
| Seedream 4 | 2K or 4K output quality selector |
Imagen (Google)
| Model | Notes |
|---|---|
| Imagen 4 | Flagship Google image model |
| Imagen Fast | Optimized for speed |
GPT-Image (OpenAI via Replicate)
| Model | Notes |
|---|---|
| GPT-Image 1.5 | Best instruction following; OpenAI API key optional (uses proxy) |
| GPT-Image 1 | Flagship; requires your own OpenAI API key |
| GPT-Image 1 Mini | Cost-efficient; requires your own OpenAI API key |
All GPT-Image models support up to 10 reference images.
Other
| Model | Notes |
|---|---|
| Recraft V4 | Vector-style and artistic outputs |
| Ideogram V3 Turbo | Fast; strong text rendering in images |
Common Controls
All models share these core controls:
| Control | Description |
|---|---|
| Prompt | Describe what you want to generate |
| Negative Prompt | Describe what to exclude from the image |
| Aspect Ratio / Size | Varies by model — some use selectors (16:9, 1:1, 9:16), others use pixel dimensions |
| Steps | Sampling iterations. Higher = more detail but slower (typical range: 20–100) |
| Guidance Scale | How strictly the model follows your prompt. Higher = more literal (typical range: 1–15) |
| Seed | Set a specific number for reproducible results; leave blank for random |
Model-Specific Controls
Output Format
For models that support it (Flux, Google, GPT-Image):
– WebP — smallest file size, good for web
– PNG — lossless, best quality
– JPEG — standard format, widely compatible
Reference Images
Some models can use reference images to guide the style or composition:
1. Under Reference Images, click Add Image.
2. Enter the URL of a publicly accessible image.
3. Add up to the model’s maximum (varies: 8–14 for most, 10 for GPT-Image).
Seedream Quality
Select 2K or 4K output resolution using the quality selector.
GPT-Image Additional Controls (requires OpenAI API key)
| Control | Options |
|---|---|
| Quality | Auto, Low, Medium, High |
| Background | Auto, Opaque, Transparent |
| Input Fidelity | Low (loose interpretation), High (strict to reference) |
AI Prompt Optimizer
Enable the Optimize Prompt toggle to have the app automatically rewrite your prompt for better results before sending it to the model. This is useful if you’re new to prompt writing.
Generating an Image
- Select your model.
- Write your prompt (and optionally a negative prompt).
- Configure size, steps, guidance, and any model-specific settings.
- Click Generate.
- The job is added to the Job Queue and runs in the background.
- When complete, the result appears in the Recent Results section and in the Results tab.
Using a Generated Image
From the Results tab or Recent Results:
– Download — saves the image file to your computer
– Add to Timeline — inserts the image as a still-frame video segment at the current playhead position
Previous: Animated Captions | Next: AI Video Generation →