What’s this?

• We will use ControlNet to further refine the interesting technique of Flux introduced here, which can generate consistent frames using only prompts.
• The method of using grids to achieve consistent images has been well-known as the sprite sheet technique since the days of Stable Diffusion 1.5. However, it seems that Flux enhances this technique even further.
• cf. Tips for Temporal Stability, while changing the video content

Workflow

• 🟩The key is the prompt. By dividing the image into four frames and specifying “In the first frame, … In the second frame, …”, we can generate consistent frames.
• 🟦I prefer using ControlNet Tile, so I don’t use Open Pose or Depth. However, the Tile in Flux seems to lack flexibility, making parameter adjustments difficult.
• The second KSampler is a simple image-to-image process.