[Feature] Text-to-Image Generation - Generate Keyframe Images via Replicate

[vzeman]

Summary

Implement text-to-image generation for creating first/last frames and reference images. High-quality keyframe images are essential for seamless scene-to-scene continuation and for image-guided video generation.

Requirements

• Supported models:
  • FLUX 1.1 Pro / FLUX Dev / FLUX Schnell
  • SDXL (Stable Diffusion XL)
  • Ideogram v3
  • Recraft v3
  • Stable Diffusion 3.5
  • DALL-E 3 (via OpenAI optional integration)
• Model-specific parameter forms (steps, CFG, aspect ratio, style)
• Generate 1, 2, or 4 image variants in one request
• Image gallery picker: select one or more generated images
• Upscaling option after generation (Real-ESRGAN or similar)
• Style presets for consistent visual language across project
• "Generate variations" of an existing image
• Inpainting: edit specific area of existing image
• Prompt history per project
• Save image to project asset library
• Image dimensions: presets matching video aspect ratios

Acceptance Criteria

• User can generate images from all supported models
• Batch generation of 4 variants works in a single request
• Generated images can be saved directly to the asset library
• Style presets produce consistent results across scenes
• Inpainting tool allows targeted area editing

Related

Part of the ai-video-editor Flutter application. Used by #64 (First & Last Frame Control) and #73 (Asset Library).