โ† Home

Images 2.6: Faster Creation, GIFs, and Precise Editing

Peter Bubenik ยท OpenAI Research ยท ยท Source

Step-by-Step Teaching

Step 1: What Is Images 2.6?

Think of Images 2.6 as an upgraded AI image engine built by OpenAI.

It is available in three places:

  • ChatGPT โ€” for everyday users
  • Codex โ€” for developers
  • The API โ€” for businesses building products

Why does it matter?

People generate 4 billion images per week using OpenAI tools. Images 2.6 is the response to that massive demand โ€” making the process faster, smarter, and more flexible.

Step 2: The Two API Model Variants

Images 2.6 comes in two distinct versions for API developers. Understanding the difference is critical.

FeatureGPT-Image-2.6 FlareGPT-Image-2.6 Sunburst
Best forSpeed, quick draftsDetail, precision
Generation speedAs fast as 6 seconds (low quality)Slower but higher fidelity
Quality settingLowMedium or High
StrengthsPosters, product imagery, layoutsCharacter consistency, brand accuracy

๐Ÿ”‘ Key Mental Model:

Flare = Fast. Sunburst = Fine.

If you are prototyping or need quick iterations โ†’ choose Flare If you are delivering a final polished asset โ†’ choose Sunburst

Step 3: Speed Improvement

Images 2.6 generates images 50% faster than the previous version.

Why this matters practically:

  • Faster iteration means more creative experiments per session
  • Reduces waiting time in production pipelines
  • Enables near real-time feedback loops for designers and developers

Step 4: The New GIF Capability

This is one of the most significant new features โ€” Images 2.6 can now generate animated GIFs from a single text prompt.

How it works conceptually:

You write a detailed prompt describing:

  1. The character and their appearance
  2. A sequence of frames (e.g., 6 frames in a 3ร—2 grid)
  3. The mood, style, and motion
  4. What should stay consistent across frames

Real Example Breakdown:

Looking at the chibi office worker prompts, notice the structure:

1. Animation format โ†’ "sprite sheet for a seamless looping GIF"
2. Character description โ†’ "chibi adult office worker, oversized round head..."
3. Frame sequence โ†’ "6 sequential frames, 3-column by 2-row grid"
4. Consistency rules โ†’ "Fixed camera, same character, desk, outfit"
5. Style โ†’ "kawaii chibi 3D toy-like, warm pastel lighting"
6. Exclusions โ†’ "No captions, no words, no logos..."

๐Ÿ”‘ Key Lesson:

Specificity drives quality. The more precisely you define frames, consistency rules, and exclusions, the better the output.

Use Cases for GIFs:

  • Brand mascot animations
  • Social media reactions and memes
  • Animated storyboards before video production
  • Typing indicators in apps (like the Town example)
  • Game characters from simple sketches

Step 5: New API Controls for Developers

Images 2.6 introduces three important parameters developers can now control:

1. ๐Ÿ–Š๏ธ Image Markup

  • Already existed in ChatGPT, now available in the API
  • Allows you to pinpoint specific areas of an image for targeted edits
  • Think of it like highlighting a sentence before editing โ€” precise, not broad

2. ๐ŸŽ›๏ธ Complexity Parameter

Controls how much planning the model does before generating

Complexity LevelBest Used For
LowQuick edits, simple changes
MediumBalanced scenes
HighScenes with many elements or constraints

Recommended default: Low complexity + your required quality level for most editing tasks

3. ๐Ÿ–ผ๏ธ Image Quality

Controls visual detail and fidelity of the output

  • Low โ†’ faster, less detailed
  • Medium/High โ†’ slower, more refined

๐Ÿ”‘ Key Distinction:

Quality = how it looks. Complexity = how hard the model thinks. These are two separate dials you can tune independently.

4. ๐Ÿ“ SVG Export

  • Outputs are now available as SVG files
  • SVG = Scalable Vector Graphics โ†’ works perfectly for websites, apps, and print without losing quality

Step 6: Practical Decision Framework

Use this when approaching a task with Images 2.6:

START
  โ”‚
  โ”œโ”€ Do you need motion? 
  โ”‚     YES โ†’ Use GIF generation with a detailed sprite sheet prompt
  โ”‚     NO  โ†’ Continue
  โ”‚
  โ”œโ”€ Is this a draft or final output?
  โ”‚     DRAFT โ†’ Flare + Low quality + Low complexity
  โ”‚     FINAL โ†’ Sunburst + Medium/High quality + adjust complexity
  โ”‚
  โ”œโ”€ Do you need to edit a specific area?
  โ”‚     YES โ†’ Use Image Markup (API)
  โ”‚     NO  โ†’ Standard generation
  โ”‚
  โ””โ”€ Does the scene have many elements?
        YES โ†’ Medium or High complexity
        NO  โ†’ Low complexity

Step 7: Availability Summary

PlatformAvailable?Notes
ChatGPTโœ… YesAll subscription tiers, desktop/mobile/web
Codexโœ… YesAll tiers
APIโœ… YesFlare and Sunburst variants, separate pricing

Quick Recap โ€” What You Now Know

ConceptKey Takeaway
Images 2.6 purposeFaster, more controllable AI image generation
Flare vs SunburstSpeed vs Quality tradeoff
Speed gain50% faster than previous version
GIF generationAnimated outputs from detailed single prompts
Complexity parameterControls model planning effort, not visual quality
Quality parameterControls visual detail and fidelity
Image MarkupPrecise area-specific editing in the API
SVG exportScalable output for professional use

Self-Check Questions

  1. You need a quick draft of a product banner. Which model variant and settings would you choose?
  2. What is the difference between the complexity and quality parameters?
  3. Write the key structural elements you would include in a GIF generation prompt.
  4. A developer wants to change only the background of an existing image. Which feature should they use?
  5. Why would a startup like Town use GIF generation rather than static images for their AI assistant?

More to study