Image-Model Prompt Adherence Test
PreregisteredThe protocol below is frozen. No outputs have been generated or scored yet, so this page makes no quality claims — it commits to how the claims will be made.
Protocol frozen 2026-08-13 · methodology v1.0 · 3 runs per prompt per model
Models under test
Metrics
| Metric | How it's scored | Scale | Better |
|---|---|---|---|
| Instruction completion | human rating | 0-10 checklist | Higher |
| Object-count accuracy | measured and rated | percent | Higher |
| Composition adherence | human rating | 1-5 | Higher |
| Requested text accuracy | measured | character accuracy rate | Higher |
| Failure rate | measured | percent | Lower |
Test protocol
- Aspect ratio
- Use the aspect ratio explicitly requested by each prompt or the nearest common supported ratio.
- Prompt enhancement
- false
- Output count
- 1
Rating & integrity
- 2 blinded raters score every output. Randomize model labels and output order independently for every prompt.
- Publish both scores, their absolute difference, and the adjudication rate.
- Count provider errors, safety refusals, corrupt media, and empty outputs as failures; do not silently rerun beyond the preregistered run count.
- Exclude only confirmed infrastructure failures affecting every compared model; record the reason and original run ID.
- Cherry-picking prohibited; no sponsored compute; no provider review before publication.
Results — not yet published
This study publishes no scores until every preregistered run is complete, both blinded raters have scored every output, and two reviewers have signed off. When that happens the summary tables and the full raw run data will appear here — including failure rates and rater disagreement, not just the winning numbers.
Frozen prompt set: image-adherence-2026-q3-v1
All 20 text-to-image prompts, frozen 2026-08-13 at version 2026-q3-v1. Every model runs every prompt 3 times. Download as JSON.
- 01Studio photograph of exactly three unbranded glass bottles: one cobalt, one amber, and one clear, arranged left to right on a white plinth; no other bottles, text, or logo.
- 02Top-down photograph of a red notebook centered between two black pencils that remain parallel to each other; pale gray desk, hard morning shadow.
- 03A square room with one moss-green chair in the back-left corner and one round white table in the front-right corner; straight architectural lines.
- 04Editorial portrait of one synthetic adult wearing a yellow raincoat and holding a closed blue umbrella in the right hand; plain gray background.
- 05Minimal poster with a black background and one centered white headline reading exactly MADE FOR MOTION; no other words, marks, or logos.
- 06Product photograph of a translucent orange sneaker facing left on a brushed aluminum plinth; warm gray background, softbox from camera right.
- 07Overhead food photograph containing exactly four blood-orange halves and six dark cherries on pale travertine; no utensils, hands, or text.
- 08Wide illustration of a lighthouse on the far right of a rocky island with a small red boat on the far left; blue hour, calm ocean.
- 09Close photograph of two silver rings placed inside a small open navy box; one ring vertical and one ring flat, soft cream background.
- 10A white ceramic robot with a square head raises its left arm while standing beside a waist-high blue cube; clean studio, full body visible.
- 11Fashion photograph of two synthetic adults wearing matching charcoal suits, standing back to back under one circular ceiling light; symmetrical framing.
- 12Flat vector icon sheet containing exactly six fruit icons in a 3 by 2 grid: apple, pear, orange, lemon, cherry, strawberry; white background.
- 13Photorealistic red bicycle leaning against a yellow wall, front wheel turned slightly right; one green plant pot at frame left and no people.
- 14Packaging mockup with one blank ivory skincare tube and one blank amber jar on a pale sage surface; soft shadows, no label text or logo.
- 15A nighttime street with three illuminated windows on the left building and two on the right building; one cyclist crossing the center.
- 16Macro photograph of one blue eye reflected in a circular silver mirror held by a hand; black background, crisp eyelashes, no extra eyes.
- 17A wooden shelf with exactly five books ordered by height from shortest on the left to tallest on the right; neutral studio light.
- 18Isometric illustration of a tiny greenhouse containing two plant beds and one central path; transparent roof, muted green palette.
- 19Commercial photograph of a clear sparkling-water can covered in condensation, centered on pink acrylic with one lime wedge behind it; no text.
- 20A four-panel comic arranged in a 2 by 2 grid showing the same small white dog finding a red ball; consistent collar and room in every panel, no captions.
Limitations
- Some models may internally rewrite prompts even when Made does not enable prompt enhancement.
- Text rendering is one component of adherence and will not be combined into a single “best model” claim without metric-specific qualification.
Disclosures
- Testing uses Made production configurations available on the stated date.
- No provider sponsors this study.
All studies in the program are listed on the research index; the models under test are priced at wholesale on the price sheet.