AI reading list
AI for creative work
Choose a creative task: describe a new image, change an existing one or assemble video clips. These authors show prompts beside results, explain how visual references guide a model, and investigate why an edit changes more than intended. Start with the practical examples; the research readings add the mechanisms behind image and video controls.
- Selected sources
- 6
People to read
Open the first material, then continue with the author on X. People appear once, grouped by their main reading use and ordered by handle within each group.
6 people
Create an image
Describe composition, subjects and details, then compare the result with your brief. Explore what a reference image can preserve through changes.
-
Turn an image idea into instructions you can check. Woolf publishes prompts beside actual outputs, including precise edits, positional ambiguity and missed requirements. His experiments give you a practical way to inspect the result.
Start with this Web material · minimaxir.com
Write a visual brief and check each change
Start with “Hello, Nano Banana!” and its five requested edits, then “Image Prompting Like an Engineer”. Compare each requirement with the image. These are his experiments with the model available in November 2025.
This work in the author’s profile
Published . Material checked .
-
Explore how a reference image can carry a character or object into a new scene. Rombach coauthored FLUX.1 Kontext; the report separates edit accuracy, visual quality and identity consistency, with examples of repeated editing.
Start with this Web material · arxiv.org
Use a reference image through successive edits
Begin with Figure 1 and the iterative examples in section 4.3, then section 5 on limitations. This team report shows artifacts that accumulate through edits; the method sections are more technical.
This work in the author’s profile
Published . Material checked .
Edit images and video
Choose what should change and what should stay. Read examples of instruction-based image editing and preserving structure across video frames.
-
Specify a change to an existing image and inspect what else moves. Brooks shares equal contribution on InstructPix2Pix, whose paired examples and failure cases make instruction-based editing concrete.
Start with this Web material · instruct-pix2pix.timothybrooks.com
Edit a picture with a written instruction
Compare the input, instruction and output, then the failure cases. Brooks and Holynski share equal contribution, with Efros also credited. The project page leads to the technical paper.
This work in the author’s profile
Material checked . Authorship source.
-
Decide which parts of a video should change: appearance, scene structure or motion across frames. Esser’s coauthored research explains these controls and why preserving structure can restrict an edit.
Start with this Web material · arxiv.org
Change a video’s appearance while preserving its structure
Start with Figures 1 and 4, then section 3.3 on content and structure. Read it as the team's 2023 research, with familiarity with diffusion helpful for the method and equations.
This work in the author’s profile
Published . Material checked .
Shape the result
Understand what a model control changes, then build an iteration loop. Adjust the output or regenerate a clip before assembling the final work.
-
See how a builder turns generated clips into an editable video. Levels describes a timeline and regenerating one clip inside Photo AI. It is a useful account of iteration from the creator of the commercial product.
Start with this Web material · levels.io
Assemble and revise generated video clips
Read the timeline and clip-regeneration workflow. Levels describes his own Photo AI product; the agent sidebar proposed at the end is a future addition in this post.
This work in the author’s profile
Published . Material checked .
-
Understand a diffusion model’s guidance setting before increasing it. Dieleman illustrates how conditioning changes the denoising path and why stronger guidance can create distortions instead of a better picture.
Start with this Web material · sander.ai
See what stronger diffusion guidance changes
Begin with “Visualising diffusion sampling” and the guidance diagrams. The visual intuition comes first; probability and machine-learning familiarity help with the equations. This control applies to the diffusion methods discussed.
This work in the author’s profile
Published . Material checked .
No people match that search. Try another name, subject or reading task.
A few bearings for your reading
These field notes explain the people, language and questions you may meet along the way.
How to write an AI image-editing brief and check the result
Define what changes, what stays and how to check an AI image edit. A published pancake example shows why those decisions matter.
Where to go next
Vibe coding & coding agents
Want to implement a media tool with a coding agent? Follow the product, coding and review workflows.
LLM research & explanations
Compare image and video controls with the architecture, training and reasoning studies behind language models.
Local models & open projects
Want to inspect the runtime, model files and hardware behind an implementation? Explore local execution and adaptation.
How this list is selected
We selected original experiments, explanations and coauthored research about images and video. Reading notes point to examples before technical sections and credit team contributions. Product-builder accounts describe their own commercial tools; research papers report methods under their stated conditions.