Versely

    AI model finder

    An AI model finder should filter by inputs, not by marketing names. Pick by what you can actually hand it. 148 models, 77 of them ranked, 9 that accept two or more reference images. Filter by input requirement rather than by name — the name is the least reliable signal there is.

    148 of 148
    ModelProviderRankNeedsRefsDurations
    GPT Image 2 Text to ImageOpenAI#1prompt——
    Gemini 3.1 Flash TTSGoogle#3prompt——
    Nano Banana 2Google#3prompt14—
    Happy Horse 1.0 Text to VideoAlibaba#3prompt—3s, 4s, 5s, 6s
    Seedance 2.0ByteDance#3video ok—4s, 5s, 6s, 7s
    Mai Image 2.5 EditMicrosoft#3image——
    Cartesia Sonic 3.5Cartesia#5prompt——
    Seedream 5.0 ProByteDance#6prompt——
    Wan 2.7 Text to VideoWan#6prompt—2s, 3s, 4s, 5s
    Happy Horse 1.1 Image to VideoAlibaba#7image—3s, 4s, 5s, 6s
    Inworld TTS 1.5 MaxInworld#7prompt——
    Grok Imagine Image QualityGrok#9prompt——
    Grok Imagine VideoGrok#9prompt—6s, 10s, 15s, 20s
    Inworld TTS 2Inworld#9prompt——
    Kling 2.5 TurboKling#9prompt—5s, 10s
    HunyuanImage 3.0 Instruct EditHunyuan#10image3—
    HiDream O1 ImageHiDream#11prompt——
    ElevenLabs MultilingualKIE#11prompt——
    Luma UNI 1 MaxLuma#11prompt——
    Flux 2 MaxFlux#12prompt——
    Vidu Q3 Image to VideoVidu#12image—5s, 10s, 15s
    Vidu Q3 VideoVidu#12prompt—5s, 10s, 15s
    Kling Image 3.0Kling#13prompt——
    Pixverse 5.6 Image to VideoPixverse#14image—5s, 8s, 10s
    Runway Gen-4.5Runway#14prompt—5s, 10s
    VEO 3.1Google#15prompt—4s, 6s, 8s
    Seedream 4.0ByteDance#16prompt——
    Pixverse 5.6 Text to VideoPixverse#16prompt—5s, 8s, 10s
    Kling Image O1Kling#16image10—
    Wan 2.7 Image to VideoWan#17image—2s, 3s, 4s, 5s
    Wan 2.7 Pro EditWan#17image——
    Pixverse V6 Text to VideoPixverse#18prompt—5s, 8s
    Kling Video V3 Standard Image to VideoKling#18image—3s, 4s, 5s, 6s
    HiDream O1 Image EditHiDream#19image——
    Flux 2 FlexFlux#19prompt——
    Pixverse V4Pixverse#20prompt—5s, 8s
    Kling Video V2.6 Pro Text to VideoKling#24prompt—5s, 10s
    Ideogram V4Ideogram#24prompt——
    Kling O3 Pro Text to VideoKling#25prompt—3s, 4s, 5s, 6s
    Krea 2 MediumKrea#26prompt——
    Wan V2.6 Text to VideoWan#28prompt—5s, 10s, 15s
    Wan 2.6 Text to ImageWan#29prompt——
    Luma Ray 2 720pLuma#29prompt—5s, 9s
    Flux 2 Klein 9B BaseFlux#29prompt4—
    Hailuo 2.3 ProHailuo#29image—6s, 10s
    Hailuo 2.3 FastHailuo#31image—6s, 10s
    Qwen Image Edit 2511Qwen#31image6—
    HiDream O1 Image DevHiDream#32prompt——
    Seedance 1.5 Pro Text to VideoByteDance#33prompt—4s, 5s, 6s, 7s
    Seedance Image to VideoByteDance#34image—2s, 3s, 4s, 5s
    ERNIE Image TurboBaidu#34prompt——
    Pixverse Text to VideoPixverse#35prompt—5s, 8s, 10s
    Recraft 4.1 Text to ImageRecraft#37prompt——
    ImagineArt 2.0Imagine#37prompt——
    Wan 2.6 Image to VideoWan#39image—5s, 10s, 15s
    Kling V2.1Kling#40image—5s, 10s
    Flux 2 FlashFlux#41prompt——
    LTX 2.3 Text to Video FastLTX#41prompt—6s, 8s, 10s, 12s
    LTX 2 Text to Video ProLTX#42prompt—6s, 8s, 10s
    LTX 2 ProLTX#43image—6s, 8s, 10s
    LTX 2 Text to Video FastLTX#43prompt—6s, 8s, 10s, 12s
    Recraft V4Recraft#45prompt——
    LTX 2LTX#45prompt—6s, 8s, 10s, 12s
    Flux 2 Klein 4B Base EditFlux#49image4—
    Midjourney V7 Image to VideoMidjourney#49image——
    PrunaAI P-VideoPrunaAI#51prompt—1s, 2s, 3s, 4s
    LTX 2.3 Image to Video ProLTX#51image—6s, 8s, 10s
    Flux KontextFlux#56image1—
    Runway Text to VideoRunway#66prompt—5s, 10s
    Imagen 4Google#70prompt——
    Qwen ImageQwen#79prompt1—
    Qwen 3 TTS 0.6BQwen#80prompt——
    Flux 1.1 ProFlux#82prompt——
    Midjourney V7Midjourney#84prompt5—
    Qwen Z ImageQwen#103prompt——
    Flux SchnellFlux#121prompt——
    Runway Gen4 ImageRunway#122image——
    Flux 3 Text to VideoBlack Forest Labs—prompt—5s, 6s, 7s, 8s
    HeyGen Avatar V3HeyGen—prompt——
    HeyGen Avatar V5HeyGen—prompt——

    Showing the first 80 of 148. Narrow the filters to see the rest.

    Rank is the model's position in Versely's overall buyer ranking; most of the catalog is unranked, which is not a mark against it — ranking only covers models compared head to head. A model without its own page is folded into a parent family in the catalog.

    Related

    FAQ

    Why filter by input instead of by name?
    Because the name almost never tells you what a model needs. A slug that sounds like an editing model can be filed as text-to-image only, with no reference support at all, and shopping by name is what produces a rejected job. Input requirements are a checkable field — whether it requires an image, accepts video, and how many reference images it takes — so that is what the filters use.
    What does the rank column mean?
    It is the model's position in Versely's overall buyer ranking. Most of the catalog is unranked, and that is not a mark against a model — ranking only covers models that have been compared head to head. An unranked model can be the right pick for a specific job.
    Why do some models have no page to click through to?
    The catalog folds variants into a parent family, so a model like an edit-mode or turbo variant is documented on its family's page rather than its own. Models whose catalog record is too thin to say anything useful also do not get a page. Both are deliberate.
    What is a reference image budget?
    The number of reference images a model will accept in one call. It matters when you need to lock more than one thing at once — a product, a face and a set, for example. A model with a budget of one cannot hold three references steady no matter how the prompt is written.