Krea AI — Image, Video & Enhancement Generation

Generate images, videos, upscale/enhance images, and train LoRA styles using the Krea.ai API. Supports 20+ image models (Flux, Imagen, GPT Image, Ideogram, Seedream...), 7 video models (Kling, Veo, Hailuo, Wan), and 3 upscalers (Topaz up to 22K).

IMPORTANT: Do NOT invent model names. Run list_models.py to get the live list of models, CU costs, and accepted parameters from the Krea API's OpenAPI spec. All scripts resolve models dynamically from the spec — there are no hardcoded endpoint tables. Scripts also accept full endpoint paths from list_models.py --json output (e.g. --model /generate/image/google/imagen-4-ultra).

Usage

Scripts are in the scripts/ directory alongside this file. Run them with uv run from the user's working directory so output files are saved where the user expects.

Generate image:
CODEBLOCK0

Generate video:
CODEBLOCK1

Enhance/upscale image:
CODEBLOCK2

Train a LoRA style:
CODEBLOCK3

List available models:
CODEBLOCK4

Run a multi-step pipeline:
CODEBLOCK5

Check job status:
CODEBLOCK6

Important: Always run from the user's current working directory so files are saved where the user is working.

Default Workflow (draft → iterate → final)

Goal: fast iteration without burning CU on expensive models until the prompt is right.

• - Draft (cheap/fast): use --model z-image or --model flux-1-dev (3-5 CU, ~5s) for quick iteration

CODEBLOCK7

• - Iterate: adjust prompt, keep trying with cheap models

• - Final (high quality): switch to --model gpt-image or INLINECODE8

CODEBLOCK8

Available Models

Models, CU costs, and accepted body fields are fetched live from the Krea API's OpenAPI spec (/openapi.json). Run list_models.py to see what's currently available:

CODEBLOCK9

Short aliases (e.g. flux for flux-1-dev) are maintained for convenience. The scripts resolve them automatically via the spec. If a model isn't in the alias list, pass the full OpenAPI model ID or endpoint path.

Model selection guidance

Map user requests for images:

• - "fast", "quick", "cheap" → flux-1-dev or INLINECODE14
• "high quality", "best" → nano-banana-pro or INLINECODE16
• "text in image", "typography" → INLINECODE17
• "photorealistic" → seedream-4 or INLINECODE19
• No preference → INLINECODE20

Map user requests for video:

• - "fast" → INLINECODE21
• "cinematic", "high quality" → INLINECODE22
• "with sound", "with audio" → veo-3 with INLINECODE24
• No preference → INLINECODE25

Enhancers: topaz-standard-enhance (faithful upscaling, default), topaz-generative-enhance (creative enhancement), topaz-bloom-enhance (adding creative details).

Image Generation Parameters

Param	Description	Default
INLINECODE29	Model ID or alias (run list_models.py)	INLINECODE30
INLINECODE31

Text description (required) | — | | --filename | Output filename (required) | — | | --width | Width in pixels (512-4096) | 1024 | | --height | Height in pixels (512-4096) | 1024 | | --aspect-ratio | Aspect ratio (1:1, 16:9, 9:16, 4:3, 3:2, etc.) | — | | --resolution | 1K, 2K, 4K (nano-banana models) | — | | --seed | Seed for reproducibility | — | | --image-url | Input image URL or local file path for image-to-image | — | | --style-id | LoRA style ID to apply | — | | --style-strength | LoRA strength (-2 to 2) | 1.0 | | --batch-size | Number of images (1-4) | 1 | | --steps | Inference steps, 1-100 (flux models) | 25 | | --guidance-scale | Guidance scale, 0-24 (flux models) | 3 | | --quality | low/medium/high/auto (gpt-image) | auto | | --output-dir | Output directory | cwd | | --api-key | Krea API token | — |

Video Generation Parameters

Param	Description	Default
INLINECODE47	Model ID or alias (run list_models.py)	INLINECODE48
INLINECODE49

Text description (required) | — | | --filename | Output filename (required) | — | | --duration | Duration in seconds | 5 | | --aspect-ratio | 16:9, 9:16, 1:1 | 16:9 | | --start-image | URL or local file path for image-to-video | — | | --end-image | End frame URL (kling only) | — | | --resolution | 720p, 1080p (veo only) | 720p | | --mode | std, pro (kling only) | std | | --generate-audio | Generate audio (veo-3 only) | false | | --output-dir | Output directory | cwd | | --api-key | Krea API token | — |

Enhancement Parameters

Param	Description	Default
INLINECODE60	Enhancer ID (run list_models.py --type enhance)	INLINECODE61
INLINECODE62

Source image URL or local file path (required) | — | | --filename | Output filename (required) | — | | --width | Target width (required) | — | | --height | Target height (required) | — | | --enhancer-model | Sub-model variant | Standard V2 | | --creativity | 1-6 (generative) or 1-9 (bloom) | — | | --face-enhancement | Enable face enhancement | false | | --sharpen | Sharpening 0-1 | — | | --denoise | Denoising 0-1 | — | | --scaling-factor | Upscaling factor 1-32 | — | | --output-format | png, jpg, webp | png | | --output-dir | Output directory | cwd | | --api-key | Krea API token | — |

LoRA Training Parameters

Param	Description	Default
INLINECODE75	Style name (required)	—
INLINECODE76

Base model: fluxdev, fluxschnell, wan, qwen, z-image | flux_dev | | --type | LoRA type: Style, Object, Character, Default | Style | | --urls | Training image URLs (space-separated) | — | | --urls-file | Text file with one URL per line | — | | --trigger-word | Trigger word to activate the LoRA in prompts | — | | --learning-rate | Learning rate | 0.0001 | | --max-train-steps | Max training steps | 1000 | | --batch-size | Training batch size | 1 | | --timeout | Polling timeout in seconds | 3600 | | --skip-validation | Skip URL HEAD-check validation | false | | --output-dir | Directory to save training manifest | — | | --api-key | Krea API token | — |

Training requires 3-2000 images. The script validates all URLs before submitting. Training takes 15-45 minutes. On completion, the style ID is printed to stdout and a training-manifest.json is saved if --output-dir is set.

Use the style ID with --style-id in generate_image.py:
CODEBLOCK10

API Key

Scripts check for API key in this order:

1. 1. --api-key argument (use if user provided key in chat)
2. INLINECODE95 environment variable

If neither is available, the script exits with an error message.

Preflight + Common Failures

Preflight:

• - command -v uv (must exist)
• INLINECODE97 (or pass --api-key)

Common failures:

• - Error: No API key → set KREA_API_TOKEN or pass INLINECODE101
• INLINECODE102 → top up compute units at https://krea.ai/settings/billing
• INLINECODE103 → model needs a paid plan upgrade at https://krea.ai/settings/billing
• INLINECODE104 → concurrent job limit reached; scripts auto-retry up to 3 times with backoff
• INLINECODE105 → check prompt for content moderation issues, try different wording

Filename Generation

Generate filenames with the pattern: INLINECODE106

• - Timestamp: current date/time in yyyy-mm-dd-hh-mm-ss (24h format)
• Name: descriptive lowercase text with hyphens (1-5 words)
• Extension: .png for images, .mp4 for videos

Examples:

• - Prompt "A cyberpunk cat" → INLINECODE110
• Prompt "waves on a beach" → INLINECODE111

Prompt Handling

For generation: Pass user's description as-is to --prompt. Only rework if clearly insufficient.

For image-to-image: Use --image-url with the source image and describe the desired transformation in --prompt.

For video from image: Use --start-image with the source image and describe the desired motion/action in --prompt.

Preserve user's creative intent in all cases.

Output

• - Scripts download the result and save it to the current directory (or --output-dir)
• Script outputs the full path to the generated file
• Do not read the image/video back — just inform the user of the saved path
• If --batch-size > 1, files are saved as name-1.png, name-2.png, etc.

Examples

Quick draft image:
CODEBLOCK11

High quality final:
CODEBLOCK12

Image-to-image edit:
CODEBLOCK13

Generate video:
CODEBLOCK14

Upscale image to 4K:
CODEBLOCK15

Train a LoRA style:
CODEBLOCK16

List models:
CODEBLOCK17

Pipelines (Multi-Step Workflows)

For multi-step workflows (generate → enhance → animate, fanout branching, template variables, parallel execution, resume, dry-run), see PIPELINES.md.

Quick example:
CODEBLOCK18