What’s new in 8.0

• A completely revamped Upscaler function, capable of generating upscaled images of higher fidelity than Magnific AI (at least, in its current incarnation) or Topaz Gigapixel.
• See the !work-in-progress! videos below (best watched in 4k).
• A new Image Enhancer function, capable of adding details to uploaded, generated, or upscaled images (similar to what Magnific AI does).
• The old Inpainter function is now split in two different functions: Inpainter without Mask, which is an img2img generation, and Inpainter with Mask function, which uses the exceptional Fooocus inpaint model to generate much better inpainted and outpainted images.
• A new Colorizer function which uses the @kijai’s DDColor node to colorize black & white pictures or recolor colored ones.
• A new Aesthetic Score Predictor function can be used to automatically choose the image of a batch that best align with the submitted prompt. The automatically selected image can then be further enhanced with other AP Workflow functions like the Hand and Face Detailers, the Object and Face Swappers, or the Upscaler.
• A new Comparer function, powered by the new, exceptionally useful Image Comparer node by @rgthree, shows the difference between the source image and the image at the end of the AP Workflow pipeline.
• The Hand Detailer function can now be configured to use the Mesh Graphormer method and the new SD1.5 Inpaint Depth Hand ControlNet model. Mesh Graphormer struggles to process hands in non-photographic images, so it’s disabled by default.
• The Caption Generator function now uses the new Moondream model instead of BLIP.
• The AI system of choice (ChatGPT or LM Studio) powering the Prompt Enricher function is now selectable from the Controller function. You can keep both active and compare the prompts generated by OpenAI models via ChatGPT against the prompts generated by open access models served by LM Studio.
• The ControlNet + Control-LoRAs function now includes six preprocessors that can be used to further improve the effect of ControlNet models. For example, you can use the new Depth Anything model to improve the effect of ControlNet Depth model.
• The Face Detailer function now has dedicated diffusion and controlnet model loading, just like the Hand Detailer and the Object Swap functions, for increased flexiblity.
• AP Workflow now supports the new version of @receyuki’s SD Parameter Generator and SD Prompt Saver nodes.
• The LoRA Keywords function is no more. The node is now part of the Caption Generator function (but it can be used at any time, independently, and even if the Caption Generator function is inactive).