GenAI Benchmarks

“I want to create merch and stickers”

T-shirt graphics, die-cut stickers and POD assets, generated at catalog scale. The dealbreakers: no real alpha channel, halo edges, and designs that fall apart at production size.

Apparel, merch & POD
Apparel & merch
The benchmark avocado mascot holding a life-size die-cut sticker of itself, next to a mug and a t-shirt with the same design
Scores

Recommended models

Same measurements as the main rankings, weighted for this job. The weight matrix is 100% covered by scored criteria (auto tier plus the interim Claude-VLM quality tier); any criterion still awaiting review is excluded and the weights renormalized, never faked.
Pilot dataset (pilot-0), weighted for apparel, merch & pod.
  Use-case score Criteria covered $ / image
1. GPT Image 1.5
Best fit
3.72 / 5 100% $0.120
2. Recraft V3 2.78 / 5 100% $0.040
3. Seedream 4.5 2.64 / 5 100% $0.048
4. Seedream 5.0 Lite 2.59 / 5 100% $0.020
5. Nano Banana 2 Lite 2.49 / 5 100% $0.020
6. Luma Photon 2.30 / 5 100% $0.019
7. FLUX.2 2.29 / 5 100% $0.013
8. Gemini 2.5 Flash Image 2.29 / 5 100% $0.039
9. Nano Banana 2 2.28 / 5 100% $0.100
10. FLUX.2 [dev] Turbo 2.28 / 5 100% $0.015
11. FLUX.2 [pro] 2.18 / 5 100% $0.040
12. Ideogram 3.0 2.08 / 5 100% $0.060
13. Qwen-Image 2.07 / 5 100% $0.030
14. Nano Banana Pro 1.96 / 5 100% $0.360
15. Stable Diffusion 1.5 1.70 / 5 100% $0.010
Want the raw numbers instead? See the full model rankings
Methodology

How this is weighted

A POD asset that arrives without a real alpha channel is not a product, it is a background-removal ticket, so transparency dominates the matrix. Resolution, prompt adherence and cost share the next tier because catalog-scale generation multiplies every per-image weakness; color, text and latency trail for graphics that are typically cleaned up or upscaled before printing anyway.

Weights are provisional (pilot-0) and published in full; the reviewed matrix ships with the frozen suite. Read the methodology
  • Transparency 35%
  • Resolution 15%
  • Prompt Adherence 15%
  • Cost 15%
  • Color Accuracy 10%
  • Text Accuracy 5%
  • Latency 5%
Evidence

The evidence: GPT Image 1.5

GPT Image 1.5 takes the top spot where this job puts its weight: transparency at 4.2 / 5 (35% of the matrix), prompt adherence at 4.6 / 5 (15% of the matrix) and resolution at 3.0 / 5 (15% of the matrix).

The best-fit model's results on the benchmarks that matter for this use case. Click through for the full cross-model comparison.
GPT Image 1.5: Mascot Logo

Mascot Logo

GPT Image 1.5: Die-Cut Sticker

Die-Cut Sticker

GPT Image 1.5: Icon Set

Icon Set

GPT Image 1.5: Monochrome Emblem

Monochrome Emblem

Takeaway

The last mile: what no model delivers

Transparency is the clearest gap in the whole benchmark: across every cut-out prompt the average alpha score is about 0.5 of 5, and only one model emits a real alpha channel. If your product promises stickers or apparel graphics, plan for background removal and edge cleanup in an editing step, not for the model to deliver a production file.

AI Editor

Generate with any model. Finish in the AI Editor.

Give your users the control they need to be productive with generative AI. The IMG.LY AI Editor turns apparel, merch & pod generations into finished, on-spec assets: background removal, brand kits and editable text on a real canvas.

Explore the AI Editor Open Demo
IMG.LY AI Editor: generative image tools next to manual editing controls on a canvas