demobook

Higgsfield: Image-to-image editing for scene consistency

Demo summary

The creator shows how to use an existing generation as a reference image to perform specific edits like adding pipes and changing lighting in Higgsfield AI.

Step-by-step

  1. Download the initial generated image to use as a reference.
  2. Use an LLM like Claude to generate a descriptive edit prompt based on the original image.
  3. Drag the reference image into the Higgsfield interface.
  4. Paste the edit prompt into the prompt field.
  5. Select the model version, such as Nanaban Pro or GPT Image 2.0.
  6. Set the batch size to four and click generate.
  7. Review the results and refine the prompt in the LLM for specific object placement or lighting adjustments.

Options

  • Generate using the Nanaban Pro model
  • Generate using the GPT Image 2.0 model
  • Adjust batch size from one to four

Watch out for

  • GPT Image 2.0 may produce results that are too dark compared to other models.

Tips

  • Work in batches of four so you can see multiple variations of the same prompt in a single row.
  • Use specific terms like 'crush the blacks' to add shadow depth and improve composition.
  • Add 'lore' or character-specific items (like a skateboard) to make the scene feel more lived-in.

Highlights

It does pretty well with edits. So, let's see how it does here.

All demos from “How to Make AI Films That Don't Look AI (Full Tutorial)

  1. 2:561:52Building a cinematic shot list with ClaudeThe creator demonstrates uploading a custom 'skill' zip file to Claude to generate a detailed shot list and style prefix for an AI film project.ClaudeAI Storyboard Generator
  2. 9:422:36Iterative environment generation with Higgsfield AIThe user demonstrates generating and refining a 3D-consistent apartment location using text prompts and the Higgsfield AI image generator.HiggsfieldAI Image Generator
  3. 19:083:46Image-to-image editing for scene consistencyCurrentThe creator shows how to use an existing generation as a reference image to perform specific edits like adding pipes and changing lighting in Higgsfield AI.HiggsfieldAI Inpainting
  4. 48:143:52Generating video from image references in Higgsfield AIThe video demonstrates the process of using character sheets and location references to generate a video clip of a character entering a room using Higgsfield AI.HiggsfieldImage to Video
  5. 2:02:332:35Troubleshooting AI video motion and FPSThe creator demonstrates how to identify and fix frame rate issues and 'morphing' artifacts in Higgsfield AI video generations by adjusting prompt parameters.HiggsfieldAI Video Generator
  6. 2:50:294:47Generating emotional character performancesThe user demonstrates prompting Higgsfield AI for specific emotional physical cues, such as a trembling lip and sobbing, to create a dramatic character breakdown.HiggsfieldText to Video
  7. Watch “How to Make AI Films That Don't Look AI (Full Tutorial)” →

AI Inpainting

  1. 5:330:54Enhance presentation slides with HiggsfieldThe video shows how to take a screenshot of a generic slide and use Higgsfield's GPT Image 2 to replace placeholders with high-quality cinematic portraits.Creating with Conor
  2. 19:083:46Image-to-image editing for scene consistencyCurrentThe creator shows how to use an existing generation as a reference image to perform specific edits like adding pipes and changing lighting in Higgsfield AI.Higgsfield AI
  3. 7:470:37Create and modify AI characters with Nano Banana ProThe video shows how to use the Nano Banana Pro model on Higgs Field to generate an AI character and then modify their clothing using specific text prompts.Mira AI
  4. 2:370:24Create custom perspectives with Angles appThe user demonstrates the 'Angles' tool by uploading an image and prompting it to rebuild the scene from a low-angle perspective while maintaining character consistency.Roboverse
  5. 6:260:23Add animals to a scene using Higgsfield Cinema StudioThe creator shows the process of using Higgsfield Cinema Studio 3.5 to add a cat and a raccoon to a specific area of a couch based on a reference photo.Joris Hermans
  6. 7:200:26Translate YouTube thumbnails with Higgsfield SupercomputerThe demo shows Higgsfield Supercomputer taking an original English YouTube thumbnail and automatically generating translated versions in French, German, and Italian.Greg Preece
  7. 7:560:31Edit character views with Nano Banana ProThe user demonstrates prompting Nano Banana Pro to change a character's orientation (e.g., 'front view') to create start and end frames for a video.Atomic Gains