rank int64 | model string | test_date string | access_route string | settings string | cost_usd string | cost_per_edit_usd string | passes int64 | partials int64 | fails int64 | output_size string | charged_total_usd string | list_total_usd string | source_url string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1 | Meta Muse Image | 2026-11-17 | WaveSpeed | source: meta/muse-image/text-to-image (aspect_ratio 3:4; output_format png); edits: meta/muse-image/edit (output_format png); first output, each edit from the untouched source, prompt only | $0.08 | $0.011 | 6 | 0 | 0 | 1344 × 1792 (3:4); edits grew slightly | 0.0770 | 0.0770 | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
2 | MiniMax H3-Image | 2026-11-17 | WaveSpeed | source: minimax/h3-image/text-to-image (aspect_ratio 4:5; resolution 2k; output_format png); edits: minimax/h3-image/edit (resolution 2k; output_format png); first output, each edit from the untouched source, prompt only | $0.26 | $0.038 | 6 | 0 | 0 | 1824 × 2304; edits softer than source | 0.2580 | 0.2580 | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
3 | FLUX 3 Image | 2026-11-17 | WaveSpeed | source: black-forest-labs/flux-3/text-to-image (aspect_ratio 4:5; resolution 2k; output_format png); edits: black-forest-labs/flux-3/image-edit (resolution 2k; output_format png); first output, each edit from the untouched source, prompt only | $0.70 | $0.10 | 6 | 0 | 0 | 1824 × 2288, kept | 0.7000 | 0.7000 | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
4 | Nano Banana 2 | 2026-07-24 | WaveSpeed | 2K where supported; first output, each edit from its own untouched source | $0.74 | $0.105 | 6 | 0 | 0 | 2K, kept | not recorded | not recorded | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
5 | Grok Imagine Image 2.0 | 2026-11-17 | WaveSpeed | source: x-ai/grok-imagine-image-v2.0/text-to-image (aspect_ratio 3:4; resolution 2k; quality medium); edits: x-ai/grok-imagine-image-v2.0/edit (resolution 2k; quality medium); first output, each edit from the untouched source, prompt only | $0.62 | $0.09 | 5 | 1 | 0 | 1776 × 2368 (3:4, JPEG), kept | 0.6200 | 0.6200 | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
6 | Qwen Image 3.0 Pro | 2026-11-17 | WaveSpeed | source: alibaba/qwen-image-3.0-pro/text-to-image (aspect_ratio 4:5; resolution 2k; enable_prompt_expansion False); edits: alibaba/qwen-image-3.0-pro/edit (resolution 2k; enable_prompt_expansion False); first output, each edit from the untouched source, prompt only | $0.54 | $0.078 | 5 | 1 | 0 | 1831 × 2289, kept | 0.5430 | 0.5430 | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
7 | Seedream 5.5 | 2026-11-17 | WaveSpeed | source: bytedance/seedream-v5.5 (aspect_ratio 4:5; resolution 2k; output_format png; background auto); edits: bytedance/seedream-v5.5/edit (resolution 2k; output_format png); first output, each edit from the untouched source, prompt only | $0.28 | $0.041 ($0.043) | 5 | 1 | 0 | 1831 × 2289; edits grew slightly | 0.2787 | 0.2930 | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
8 | Seedream 5.0 Pro | 2026-07-24 | WaveSpeed | 2K where supported; first output, each edit from its own untouched source | $0.63 | $0.09 | 5 | 1 | 0 | 2K, kept | not recorded | not recorded | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
9 | Krea 3 | 2026-11-17 | WaveSpeed | source: krea/v3/text-to-image (model berry; reasoning_level medium; size 1632*2048; output_format png); edits: krea/v3/edit (model berry; reasoning_level medium; size 1632*2048; output_format png); first output, each edit from the untouched source, prompt only | $1.24 | $0.191 ($0.21) | 5 | 1 | 0 | 1632 × 2048; edits softer than source | 1.2434 | 1.3700 | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
10 | Nano Banana 2.1 | 2026-11-17 | WaveSpeed | source: google/nano-banana-2.1/text-to-image (resolution 2k; output_format png; enable_web_search False; enable_image_search False; aspect_ratio 4:5); edits: google/nano-banana-2.1/edit (resolution 2k; output_format png; enable_web_search False; enable_image_search False); first output, each edit from the untouched sou... | $0.39 | $0.056 ($0.062) | 5 | 0 | 1 | 1856 × 2304, kept | 0.3888 | 0.4320 | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
11 | GPT Image 2.6 Sunburst | 2026-11-17 | WaveSpeed | source: openai/gpt-image-2.6-sunburst/text-to-image (resolution 2k; quality max; complexity high; output_format png; aspect_ratio 4:5); edits: openai/gpt-image-2.6-sunburst/edit (resolution 2k; quality max; complexity high; output_format png); first output, each edit from the untouched source, prompt only | $3.04 | $0.437 ($0.485) | 4 | 2 | 0 | 1712 × 2144, kept | 3.0420 | 3.3800 | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
12 | Nano Banana Pro | 2026-07-24 | WaveSpeed | 2K where supported; first output, each edit from its own untouched source | $0.98 | $0.14 | 4 | 2 | 0 | 2K, kept | not recorded | not recorded | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
13 | GPT Image 2.6 Flare | 2026-11-17 | WaveSpeed | source: openai/gpt-image-2.6-flare/text-to-image (resolution 2k; quality max; complexity high; output_format png; aspect_ratio 4:5); edits: openai/gpt-image-2.6-flare/edit (resolution 2k; quality max; complexity high; output_format png); first output, each edit from the untouched source, prompt only | $3.04 | $0.437 ($0.485) | 3 | 3 | 0 | 1712 × 2144, kept | 3.0420 | 3.3800 | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
14 | GPT Image 2.5 Sunburst | 2026-11-17 | WaveSpeed | source: openai/gpt-image-2.5-sunburst/text-to-image (resolution 2k; quality max; output_format png; aspect_ratio 4:5); edits: openai/gpt-image-2.5-sunburst/edit (resolution 2k; quality max; output_format png); first output, each edit from the untouched source, prompt only | $4.29 | $0.615 | 3 | 3 | 0 | 1712 × 2144, kept | 4.2900 | 4.2900 | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
15 | GPT Image 2.5 Flare | 2026-11-17 | WaveSpeed | source: openai/gpt-image-2.5-flare/text-to-image (resolution 2k; quality max; output_format png; aspect_ratio 4:5); edits: openai/gpt-image-2.5-flare/edit (resolution 2k; quality max; output_format png); first output, each edit from the untouched source, prompt only | $4.29 | $0.615 | 3 | 3 | 0 | 1712 × 2144, kept | 4.2900 | 4.2900 | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
16 | Qwen Image 2.0 Pro | 2026-07-24 | WaveSpeed | 2K where supported; first output, each edit from its own untouched source | $0.49 | $0.07 | 3 | 2 | 1 | Shrank to 928 × 1152 | not recorded | not recorded | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
17 | GPT Image 2 | 2026-07-24 | WaveSpeed | 2K where supported; first output, each edit from its own untouched source | $0.76 / $2.86 | $0.11 medium, $0.41 high | 2 | 3 | 1 | Shrank to 1792 × 2240 | not recorded | not recorded | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
18 | FLUX.2 Max | 2026-07-24 | WaveSpeed | 2K where supported; first output, each edit from its own untouched source | $0.49 | $0.07 | 2 | 3 | 1 | Shrank to 928 × 1152 | not recorded | not recorded | https://theseguysknow.io/best-ai-image-generator-editor-test/ |
TGK AI Image Editors
Article: https://theseguysknow.io/best-ai-image-generator-editor-test/ (updated 17 November 2026)
Version: v2 (17 November 2026). Zenodo DOI for this version: https://doi.org/10.5281/zenodo.24322904. v1: https://doi.org/10.5281/zenodo.24316573.
Eighteen editors, six edit tasks and 108 graded edits. Six editors were tested on 24 July 2026 and twelve on 17 November 2026, all through WaveSpeed. Every editor generated its own café source photo from the same prompt, then made six separate edits from that untouched source, using the prompt only. We kept the first output of every job. Midjourney V8.1 took part in July as a generation-only model and is excluded from the edit ranking.
The image inventory lists 135 assets: 19 source generations (seven from July, including Midjourney V8.1, and twelve from November), 108 edit outputs, seven July feature images and the November feature image. Images are linked, not included.
Results
| rank | model | test_date | passes / partials / fails | full test | per edit |
|---|---|---|---|---|---|
| 1 | Meta Muse Image | 2026-11-17 | 6 / 0 / 0 | $0.08 | $0.011 |
| 2 | MiniMax H3-Image | 2026-11-17 | 6 / 0 / 0 | $0.26 | $0.038 |
| 3 | FLUX 3 Image | 2026-11-17 | 6 / 0 / 0 | $0.70 | $0.10 |
| 4 | Nano Banana 2 | 2026-07-24 | 6 / 0 / 0 | $0.74 | $0.105 |
| 5 | Grok Imagine Image 2.0 | 2026-11-17 | 5 / 1 / 0 | $0.62 | $0.09 |
| 6 | Qwen Image 3.0 Pro | 2026-11-17 | 5 / 1 / 0 | $0.54 | $0.078 |
| 7 | Seedream 5.5 | 2026-11-17 | 5 / 1 / 0 | $0.28 | $0.041 ($0.043) |
| 8 | Seedream 5.0 Pro | 2026-07-24 | 5 / 1 / 0 | $0.63 | $0.09 |
| 9 | Krea 3 | 2026-11-17 | 5 / 1 / 0 | $1.24 | $0.191 ($0.21) |
| 10 | Nano Banana 2.1 | 2026-11-17 | 5 / 0 / 1 | $0.39 | $0.056 ($0.062) |
| 11 | GPT Image 2.6 Sunburst | 2026-11-17 | 4 / 2 / 0 | $3.04 | $0.437 ($0.485) |
| 12 | Nano Banana Pro | 2026-07-24 | 4 / 2 / 0 | $0.98 | $0.14 |
| 13 | GPT Image 2.6 Flare | 2026-11-17 | 3 / 3 / 0 | $3.04 | $0.437 ($0.485) |
| 14 | GPT Image 2.5 Sunburst | 2026-11-17 | 3 / 3 / 0 | $4.29 | $0.615 |
| 15 | GPT Image 2.5 Flare | 2026-11-17 | 3 / 3 / 0 | $4.29 | $0.615 |
| 16 | Qwen Image 2.0 Pro | 2026-07-24 | 3 / 2 / 1 | $0.49 | $0.07 |
| 17 | GPT Image 2 | 2026-07-24 | 2 / 3 / 1 | $0.76 / $2.86 | $0.11 medium, $0.41 high |
| 18 | FLUX.2 Max | 2026-07-24 | 2 / 3 / 1 | $0.49 | $0.07 |
Costs are what we paid on the test date (July list prices; November charges from WaveSpeed billing, with the list price in brackets where a discount applied). They are not current price quotes. The November full test covers the source and six edits.
Method and limitations
- November runs used 2K where offered and the highest practical quality: max for the GPT Image models and medium for Grok, its highest. Meta Muse Image and Grok used 3:4 because neither offers 4:5. Prompt expansion was off wherever the form allowed it.
- Every November job is in evidence_records.csv with its endpoint, settings, output size, list price, charge and WaveSpeed prediction ID.
- The verbatim November prompts are in test_prompts.csv. The July edit wording was not preserved; the July requests match the six summaries.
- The GPT Image 2 quality setting used in July is not established.
- Published images are WebP or PNG copies of the recorded outputs and may be resized.
- Six edits on one scene don't cover every editing job. Masks, inpainting brushes and reference images were not tested.
Files
model_results.csv: 18 editors with rank, test date, settings, costs, grade counts and output size.evidence_records.csv: 108 graded edits with the grade, the reason for each Partial or Fail, image URL and run details.image_inventory.csv: every source, edit and feature image with its published caption.test_prompts.csv: the source prompt and six edit prompts.SCORING.md,DATA_DICTIONARY.md,CHANGELOG.md,CITATION.bib,LICENSE.
Dataset mirrors and citation
Hugging Face: https://huggingface.co/datasets/These-Guys-Know/tgk-ai-image-editors Kaggle: https://www.kaggle.com/datasets/mikehazard/tgk-ai-image-editors Zenodo (this version): https://doi.org/10.5281/zenodo.24322904 Zenodo (v1): https://doi.org/10.5281/zenodo.24316573
- Downloads last month
- 56