AI Models
The models behind Visxy
What each model is good at, where it differs from the others, and what it costs to run.
AI Models
What each model is good at, where it differs from the others, and what it costs to run.
AI Models
What each model is good at, where it differs from the others, and what it costs to run.
Google's newer Gemini image model, an update of Nano Banana 2. Sharper pictures and type, still a prompt plus up to four pictures you name.
Kuaishou's Kling 4.0. A 3 to 30 second clip in 720p or 1080p, from a prompt or from a sequence of frames.
Google's studio Gemini image model. Use it when type, a dense layout, or a product has to stay exact, not while you are still trying ideas.
Kling 4.0 Flash is the shorter Kling 4.0 model. A 3 to 20 second clip at 720p, from a prompt or one opening frame. Reference pictures and reference clips are separate paths.
OpenAI's faster GPT Image 2.6 model. More frame sizes, a quality choice from Low through Max, and edits of up to four pictures.
OpenAI's higher-detail GPT Image 2.6 model. The same sizes and quality choices as Flare, for edits that have to stay consistent.
ByteDance's Seedance 2.5. A clip that can hold a whole beat of story, from a prompt or from an opening frame and an optional closing frame.
OpenAI's lower-cost GPT Image 2.6 model. Still pictures at 1K, or an edit of up to four pictures you name as image 1.
Kuaishou's Kling 3.0 video model. A clip from a prompt or from an opening frame and an optional closing frame, with sound you can turn on.
ByteDance's Seedream 5.5. A still at 1K, 1.5K, or 2K, or an edit of up to four pictures. The edit's price follows the size and how many pictures you attach.
Black Forest Labs' higher Flux 3 video tier. A 5 to 20 second clip at 720p or 1080p, from a prompt or from a start frame and an optional end frame. Sound stays off until you turn it on.
ByteDance's Seedream model for designed pictures: dense layouts, readable type, and realistic materials. Attach up to four pictures and name them.
Google's Veo 3.1 video model. A short clip with sound from a prompt or from one opening picture. Fast and Lite are not offered.
Black Forest Labs' production FLUX.2 stills model. Photoreal product, material, and light, or an edit of up to four pictures you name.
Google's Gemini Omni 1.2 Flash video model. A 3 to 15 second clip with sound, at up to 4K, from a prompt, frames, or two to six reference pictures.
Alibaba's faster Wan 3.0. The same prompt, frames, sound switch, and reference pictures, aimed at a quicker clip.
Alibaba's Qwen Image 3 Pro. The Pro model beside Qwen Image 3: a dense picture from a prompt, or an edit of one picture.
Black Forest Labs' first Flux video model. A clip with sound from a prompt or from one opening picture. Flux 2 still makes the photographs.
OpenAI's faster GPT Image 2.5 model. Everyday pictures and edits, with up to four pictures you name in the prompt.
MiniMax H3 Max Turbo is the faster H3 Max tier. A clip of 1 to 15 seconds with sound, at 480p, 768p, or 1080p, from a prompt or from a start and an optional end frame.
OpenAI's higher-precision GPT Image 2.5 model. For edits and pictures that have to hold detail. Attach up to four pictures you name.
MiniMax H3 Max makes a clip of 1 to 15 seconds with sound, at 480p, 768p, or 1080p, from a prompt or from a start frame and an optional end frame.
ByteDance's fast Seedream 5 image model. A picture from a prompt, or an edit of one to four pictures, without @ tags.
xAI's Grok Imagine 1.5 video model. A clip from a prompt or one opening picture, or from several reference pictures without @ tags.
Alibaba's Qwen Image 3. A dense picture from a prompt, or an edit of one picture. Enhance expands the prompt only if you turn it on.
xAI's lighter Grok Imagine 1.5 video model. A clip of 1 to 15 seconds at 480p, 720p, or 1080p, from a prompt or one opening picture.
Ideogram 4.5 makes a still, or edits up to four pictures. Choose a quality, and turn Enhance on when you want the prompt expanded.
Change a Kling-length clip you already have. Attach one clip of 3 to 30 seconds, and up to ten pictures. Name the clip @Video1 and the pictures @Image1.
Black Forest Labs' FLUX.3 Klein. A still from a prompt, or an edit of up to four pictures, at 1K, 2K, or 4K.
Change a clip with Gemini Omni 1.2 Flash. Attach one clip of up to 15 seconds. The result keeps that length, and you pick the resolution, up to 4K.
xAI's Grok Imagine image model at the Quality tier. A still at 1K or 2K, or an edit of up to three pictures.
PixVerse V6 Max. A 1 to 15 second clip from a prompt or one opening picture, at 360p to 1080p, with sound you can turn on.
Ideogram's image model for readable type, posters, and photoreal frames. Pick a rendering speed, and remix one picture when you already have it.
Alibaba's Wan 3.0 video model. A longer clip from a prompt or two frames, with optional sound, or from reference pictures you name.
xAI's current Grok image model. A picture from a prompt, or an edit of one to four pictures on Grok Imagine 2.0 Edit.
ByteDance's Seedance 2.0. A 4 to 15 second clip at 480p, 720p, or 1080p, from a prompt or from an opening frame and an optional closing frame.
Krea 3 makes a still from a prompt, in one of six frame sizes. There is no editor. The price follows the size you pick.
The fast tier of ByteDance's Seedance 2.0. A prompt is always required, plus an opening frame and an optional closing frame. Sound starts off.
Alibaba's Qwen Image 3 as Higgsfield serves it. A still at 1K or 2K, ten frames, and an edit of up to three pictures. Enhance is off until you turn it on.
Kuaishou's Omni Kling. A clip from a prompt or two frames, a clip from several reference pictures, or a restyle of a clip you already have.
Recraft V4.1 Pro. An image from a prompt, at 2k, with no picture to attach.
Alibaba's HappyHorse 1.1. A clip from a prompt or one opening picture, or from reference pictures you name. Audio is part of the model.
Recraft V4.1. An image from a prompt, at 1k, with no picture to attach.
PixVerse V6. A clip from a prompt or one opening picture, with sound you can turn on, or a clip guided by several reference pictures.
Alibaba's sharper Wan 2.7 stills model. Steadier composition and a closer reading of the prompt. Edit one to four pictures.
MiniMax H3 Max Turbo Extend continues a clip you already have. You choose 1 to 15 seconds of new footage, at 480p, 768p, 1080p, or 2K. The clip you attach is not what you pay for.
Higgsfield's Soul 2. An image from a prompt, at 720p or 1080p. A picture you attach is edited instead.
MiniMax H3 Max Recast replaces the people in a clip with the people in your photos. The clip keeps its motion, camera, and length. Attach one photo per new person.
Higgsfield's Genjutsu motion transfer. One clip's motion, carried onto one to eight pictured subjects. A prompt is optional.
Higgsfield's Genjutsu Lite motion transfer. A 4 to 15 second clip, one to eight pictures, at 480p, 720p, or 1080p. The source audio stays.
Higgsfield's Genjutsu object swap. One clip, plus one to eight pictures of what should replace something in it. A prompt is optional.
Higgsfield's Genjutsu Lite object swap. One 4 to 15 second clip, plus one to eight pictures of what should replace something in it.
Higgsfield's Genjutsu Restyle. A new look for one clip, with its motion and audio kept. A style is required.
HeyGen makes a clip of 5 to 15 seconds with sound, at up to 2K, from a prompt, one opening picture, or two to nine reference pictures.
Black Forest Labs' FLUX.2 [flex] stills model, labeled Flux 2. The same kind of photograph as Flux 2 Pro, with their control-oriented model.
Google's general Gemini image model. A finished picture from a prompt you already trust, or an edit of up to four pictures you name in that prompt.
xAI's unversioned Grok video model. A clip from a prompt or from one opening picture. It is not the 1.5 model and it is not a still.
Google's fastest Gemini image model. Try a picture here, then move a keeper to Nano Banana 2 or Pro when the frame has to hold up.
Alibaba's Wan 2.7 video model. A clip from a prompt or two frames, or from reference pictures you name. Enhance is optional. There is no sound switch.
ByteDance's earlier Seedream 5 model, built to read a vague brief and still land the picture. Quality runs basic, high, or ultra.
Kuaishou's Kling 3.0 Turbo. A short clip from a prompt or from one opening picture. It is not a speed switch on Kling 3.0.
Lightricks' LTX-2.5 Pro. A 6, 8, or 10 second clip at 720p or 1080p, from a prompt or from a picture.
MiniMax's general video model. A clip from a prompt, from an opening and optional closing frame, or from several reference pictures.
Alibaba's earlier HappyHorse. A clip from a prompt or one opening picture, or from reference pictures you name in the prompt.
Alibaba's Wan 2.7 image model. A new picture from a prompt, or an edit of one to four pictures. Wan 2.7 Image Pro is the sharper sibling.
xAI's earlier Grok image model. Speed or quality, two pictures each time, and an editor that takes up to four pictures you can name.
Kuaishou's previous Kling. A clip from a prompt or one opening picture, with sound you can turn on. Kling 3.0 is the current model.
ByteDance's smaller Seedance model. A clip from a prompt or from an opening frame and an optional closing frame, with sound that starts off.
OpenAI's image model for layouts, readable type, and precise edits. Attach up to four pictures and name them in the prompt.
Lightricks' LTX-2.5 Fast. A 6, 8, or 10 second clip at 720p, 1080p, 2k, or 4k.
Alibaba's Qwen Image 2.1. A new picture or an edit of up to four pictures, with Enhance if you want the prompt expanded.
Luma's Ray video model on Visxy. A clip from a prompt, or from an opening frame and an optional closing frame. No sound switch.
Alibaba's Qwen-Image 2.0. One model for a new picture or an edit of up to three pictures, including long and multilingual type.
Shengshu's Vidu Q4. A clip that starts from a picture, with optional sound, or from many reference pictures you name in the prompt.
OpenAI's earlier GPT Image model. Text rendering and photoreal pictures, a medium or high quality choice, and edits of up to four pictures.
Ideogram's character model. One picture of a character, plus a prompt for the new scene. Rendering speed and Enhance work as on Ideogram V3.
MiniMax H3 Max in a VHS look. A 5 to 15 second clip with sound, from a prompt or one opening frame, with tape damage set to light, medium, or heavy.
MiniMax H3 Max in a 1970s cartoon style. A 5 to 15 second clip with sound, from a prompt or one opening frame. The style is applied for you.
MiniMax H3 Max in a low-poly style. A 5 to 15 second clip with sound, from a prompt or one opening frame. The style is applied for you.
MiniMax H3 Max in a hand-drawn style. A 5 to 15 second clip with sound, from a prompt or one opening frame. The style is applied for you.
MiniMax H3 Max in a 16-bit pixel style. A 5 to 15 second clip with sound, from a prompt or one opening frame. The style is applied for you.
Alibaba's HappyHorse 1.0 video editor. Change one 3 to 15 second clip, and guide it with up to five pictures. The result keeps the clip's length.
Kuaishou's Kling 3.0 motion transfer. One picture of a subject plus one clip whose motion it copies. A prompt is optional.
Kuaishou's Kling 2.6 motion transfer. One picture of a subject plus one clip whose motion it copies. A prompt is optional.
Kling 3.0 Omni Transformation. One clip in, a transformed clip out, with a prompt. Sound starts off.
Google's Imagen 4 Ultra. Text to image, for the strictest match to the prompt and the highest level of detail. It does not edit.
Google's Imagen 4 text-to-image model. Fine detail and readable type, from a prompt alone. It does not take a picture to edit.
Google's fast Imagen 4. A text-to-image picture when you want the idea quickly. It does not take a picture to edit.
Recraft V4.1 Utility Pro. The utility tier at 2k. A prompt, and no picture to attach.
Recraft V4.1 Utility. The utility tier at 1k. A prompt, and no picture to attach.
Higgsfield's first Soul model. An image from a prompt, at 720p or 1080p, with no picture to attach.
Tongyi's efficient image model. Photoreal pictures and bilingual type from a prompt. On Visxy it does not take a picture to edit.
The first Qwen image model from Alibaba. Strong type in the picture, plus one attached picture to edit or to transform.
Qwen Image changes one picture you attach, using your prompt. It does not start from text alone.
Topaz image upscale. One picture in, a cleaner or larger picture out. Choose a factor of 1, 2, or 4. There is no prompt.
Recraft Crisp Upscale. One picture in, a larger picture out. Recraft chooses the scale. There is no prompt.
Recraft Remove Background. One picture in, the same picture with the background removed. There is no prompt.
Topaz video upscale. One clip in, a cleaner or larger clip out, as long as the one you gave it. Choose a factor of 1, 2, or 4.