Image prompts
Each image model has its own idea of a good prompt. Find the model below, read its guide in references/, and write the prompt the way that guide describes. If you switch models, switch guides.
Model guides
Model (create_image, edit_image) |
Guide |
gpt-image-2 |
references/openai-gpt-image-2.md |
nano-banana (default), nano-banana-edit, nano-banana-2, nano-banana-2.1, nano-banana-pro |
references/google-gemini-image.md |
seedream-5-lite, seedream-5-pro, seedream-4 |
references/bytedance-seedream.md |
grok-imagine, grok-imagine-quality |
references/xai-grok-imagine.md (image sections) |
Models without a guide here (ideogram-4, ideogram-45) follow the defaults below.
In Pika
- The prompt goes in the tool's
prompt field (instructions in edit_image). A template or preset field may replace it; send one or the other.
- Keep the user's intent and words. Expand a described look into the concrete detail the guide asks for; don't swap in a different idea.
- Aspect ratio, resolution, quality, transparent background and output format are fields. Keep them out of the prompt.
- References go in
image_urls (reference_urls in edit_image); name them in that order ("image 1", "image 2") and give each one a job.
- A guide may mention a control the tool doesn't expose, such as a negative prompt or a seed. Restate a negative as the wanted result inside the prompt, or leave it out.
Defaults when a guide is silent
- Write prose, not keyword lists, and leave out empty praise ("stunning, masterpiece, 8K").
- Describe what you want, not what to avoid ("an empty street", not "no cars").
- Be concrete: subject, composition, light, style, materials.
- Put on-image text in quotes, with its placement and lettering.
- For edits, name the one change and list what stays: face, pose, product shape, label, background, camera angle. Make one change per run.
1---2name: prompt-image3description: Use when writing or improving the prompt for a Pika image run — create, edit or restyle an image — before running it. Routes each model to its maker's own prompt guide.4metadata:5 source: https://skillmd.com/skills/pika-labs/prompt-image6---78# Image prompts910Each image model has its own idea of a good prompt. Find the model below, read its guide in `references/`, and write the prompt the way that guide describes. If you switch models, switch guides.1112## Model guides1314| Model (`create_image`, `edit_image`) | Guide |15| --- | --- |16| `gpt-image-2` | `references/openai-gpt-image-2.md` |17| `nano-banana` (default), `nano-banana-edit`, `nano-banana-2`, `nano-banana-2.1`, `nano-banana-pro` | `references/google-gemini-image.md` |18| `seedream-5-lite`, `seedream-5-pro`, `seedream-4` | `references/bytedance-seedream.md` |19| `grok-imagine`, `grok-imagine-quality` | `references/xai-grok-imagine.md` (image sections) |2021Models without a guide here (`ideogram-4`, `ideogram-45`) follow the defaults below.2223## In Pika2425- The prompt goes in the tool's `prompt` field (`instructions` in `edit_image`). A template or preset field may replace it; send one or the other.26- Keep the user's intent and words. Expand a described look into the concrete detail the guide asks for; don't swap in a different idea.27- Aspect ratio, resolution, quality, transparent background and output format are fields. Keep them out of the prompt.28- References go in `image_urls` (`reference_urls` in `edit_image`); name them in that order ("image 1", "image 2") and give each one a job.29- A guide may mention a control the tool doesn't expose, such as a negative prompt or a seed. Restate a negative as the wanted result inside the prompt, or leave it out.3031## Defaults when a guide is silent3233- Write prose, not keyword lists, and leave out empty praise ("stunning, masterpiece, 8K").34- Describe what you want, not what to avoid ("an empty street", not "no cars").35- Be concrete: subject, composition, light, style, materials.36- Put on-image text in quotes, with its placement and lettering.37- For edits, name the one change and list what stays: face, pose, product shape, label, background, camera angle. Make one change per run.