GPT Image 2 Agent
An AI-orchestrated multi-frame generator: GPT 5.5 reads your brief, plans a distinct prompt for every scene, and GPT Image 2 renders each frame. One run, up to 10 coherent images — built for storyboards, shot sequences, and campaign sets.
Capabilities
| Feature | Support |
|---|---|
| Text-to-Image (multi-frame) | Yes |
| Image-to-Image | Yes (references carry into every frame) |
| Frames per Run | Up to 10 |
| Max Resolution | 4K (3840 x 3840) |
| Reference Images | Up to 6 |
| Aspect Ratios | 1:1, 1:3, 2:3, 3:1, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9 |
| Quality Tiers | 1K, 2K, 4K, Auto |
| Reasoning Effort | Low / Medium / High / X-High |
| Typical Runtime | 2–30 min (scales with frame count and effort) |
How It Works
- You write one brief — the whole sequence in a single prompt: "6-shot storyboard of a heist: exterior establishing, vault door, laser grid, the grab, alarm, rooftop escape."
- GPT 5.5 plans — it splits the brief into per-frame prompts, keeping characters, wardrobe, palette, and camera language consistent across frames.
- GPT Image 2 renders each planned frame at your chosen quality tier.
The result lands as a set on the node's output rail — each frame is a normal image artifact you can edit, reference, or feed into a video model.
Reasoning Effort
Controls how hard the planner thinks before rendering:
- Low / Medium — fine for loose mood sets and variations on a theme.
- High / X-High — worth it for narrative sequences where continuity errors (a jacket changing color between shots) are expensive. Higher effort adds planning time, not render time per frame.
Prompting Tips
- Number your shots. Explicit shot lists ("Shot 1: … Shot 2: …") give the planner clean boundaries; it fills in consistent style connective tissue.
- Pin the constants once. State character/wardrobe/palette invariants at the top of the brief — the planner propagates them to every frame prompt.
- Reference images apply globally. Up to 6 references inform all frames — great for character sheets and style bibles, not for per-frame references.
- Budget time accordingly. A 10-frame, high-effort, 4K run can take up to 30 minutes. Iterate at 1K with fewer frames first.
Limitations
- Frames are planned in one pass — you can't revise frame 3 alone and keep the plan (re-run, or edit the frame with a normal image model)
- References are shared across all frames, not assignable per frame
- No seed; runs are not reproducible
- Long runtimes at high frame counts / effort / resolution
See Also
- GPT Image 2 — the single-frame renderer underneath
- GPT 5.5 — the planner model, usable directly as a chat node