Knowledge
AI models for images and video.
Which models exist, who makes them, and what they take in and put out. All details come from the makers’ own pages and documentation and are linked to the source. We don’t sell access to these models and aren’t paid for listing them.
Kept up to date
Models, limits and prices change often. That’s why we regularly check this overview against the makers’ own pages: new models are added, discontinued ones removed and changed details updated. Every card links to its source.
- models
- 35
- makers
- 14
Image
Models for images.
Create images from a description, many also from reference images or by editing an existing image.
Adobe · United States
Firefly Image Model 5
Photorealistic images, portraits and groups, editing by description; according to Adobe trained on Adobe Stock, public-domain and licensed content.
- Inputs:
- Text, reference image, style and composition references
- Output:
- 4 megapixels
- Access:
- Adobe Firefly (web) and Adobe apps
- Open weights:
- no
Alibaba (Qwen) · China
Qwen-Image 3.0 / 3.0 Pro
Complex layouts and small text in several languages; generation and editing in one model.
- Inputs:
- Text, 1 to 3 images to edit
- Output:
- up to 2048 × 2048 px; up to 6 images per request
- Access:
- Qwen Chat, Alibaba Cloud Model Studio (API)
- Open weights:
- no
Alibaba (Tongyi) · China
Z-Image Turbo
Small 6-billion-parameter model that renders Chinese and English text; according to the maker it runs on 16 GB of graphics memory.
- Inputs:
- Text
- Output:
- up to 2048 × 2048 px (Model Studio)
- Access:
- Weights on Hugging Face (Apache-2.0), Alibaba Cloud Model Studio (API)
- Open weights:
- yes
- Released:
- November 2025
Alibaba (Wan) · China
Wan 2.7 Image / Image Pro
Generates and edits images, including several at once; editing within marked areas, brand colours and image sets with consistent characters.
- Inputs:
- Text, images to edit (Pro up to 9), marked areas
- Output:
- Pro up to 4096 × 4096 px, otherwise up to 2048 × 2048
- Access:
- Alibaba Cloud Model Studio (API)
- Open weights:
- no
- Released:
- April 2026
Black Forest Labs · Germany
FLUX 3 Image Klein
Small open-weight model (5.6 billion parameters) for generation, editing and combining several references.
- Inputs:
- Text, up to 10 reference images, marked areas
- Output:
- 1 image per request; resolution not stated
- Access:
- Weights on Hugging Face (FLUX Kommunity License; limited commercial use)
- Open weights:
- yes
- Released:
- October 2026
Black Forest Labs · Germany
FLUX 3 Image
Generates and edits images; layout controllable through marked areas, several references can be combined.
- Inputs:
- Text, up to 10 reference images, marked areas
- Output:
- 2K and 4K
- Access:
- BFL API; weights for self-hosting only under a commercial licence
- Open weights:
- no
ByteDance · China
Seedream 5.0 Lite
Image model for generation and editing that can look up current topics online.
- Inputs:
- Text, images to edit
- Output:
- resolution not stated
- Access:
- Dola, BytePlus (API)
- Open weights:
- no
ByteDance · China
Seedream 5.0 Pro
Image model with reasoning, for text-heavy layouts and infographics; edits from sketches and annotations, including layer separation.
- Inputs:
- Text, reference images, sketches, annotations
- Output:
- resolution not stated
- Access:
- Dola, BytePlus ModelArk (API)
- Open weights:
- no
Google · United States
Nano Banana 2.1
Generates and edits images conversationally, with adjustable thinking levels and optional Google Search grounding.
- Inputs:
- Text, up to 14 reference images, video
- Output:
- 1K, 2K or 4K; invisible watermark (SynthID)
- Access:
- Gemini API, Google AI Studio
- Open weights:
- no
- Released:
- October 2026
Google · United States
Nano Banana 2 Lite
Lowest-cost variant of the Nano Banana line; according to Google less suited to many references and multi-step editing.
- Inputs:
- Text, reference images
- Output:
- 1K; invisible watermark (SynthID)
- Access:
- Gemini API
- Open weights:
- no
- Released:
- June 2026
Google · United States
Nano Banana Pro
Larger Nano Banana model for more demanding tasks; can output interleaved text and images.
- Inputs:
- Text, reference images for objects and style
- Output:
- 1K, 2K or 4K; invisible watermark (SynthID)
- Access:
- Gemini API
- Open weights:
- no
- Released:
- November 2025
Ideogram · Canada
Ideogram 4.5
Generates images and edits precisely within a mask; background can be transparent.
- Inputs:
- Text or JSON, up to 5 reference images, mask
- Output:
- 1K or 2K; transparent background possible
- Access:
- Ideogram (web), Ideogram API
- Open weights:
- no
Kuaishou (Kling AI) · China
Kling Image 3.0
Kling’s image model, creating image series in a consistent style.
- Inputs:
- Text, images
- Output:
- 4K
- Access:
- Kling AI (web, app), Kling API
- Open weights:
- no
- Released:
- February 2026
Midjourney · United States
Midjourney V8.2
Midjourney’s default model, focused on aesthetics and personalised style; editing through the Edit model.
- Inputs:
- Text, image prompts, style references
- Output:
- HD up to 2048 px; aspect ratio up to 14:1
- Access:
- Midjourney web app, Discord; no API mentioned
- Open weights:
- no
- Released:
- July 2026
OpenAI · United States
GPT Image 2.5 Flare
Faster GPT Image 2.5 variant for everyday images, with editing of selected image areas.
- Inputs:
- Text, images
- Output:
- up to 3840 px edge length; PNG, JPEG, WebP
- Access:
- OpenAI API
- Open weights:
- no
- Released:
- September 2026
OpenAI · United States
GPT Image 2.5 Sunburst
Generates and edits images; designed for workflows where precise editing matters.
- Inputs:
- Text, images
- Output:
- up to 3840 px edge length; PNG, JPEG, WebP
- Access:
- ChatGPT, OpenAI API
- Open weights:
- no
- Released:
- September 2026
Recraft · United States
Recraft V4.1
Generates raster images and true vector graphics (SVG) with custom styles, plus variants for mockups and product shots.
- Inputs:
- Text, 1 to 6 images, up to 10 style references, colour preferences
- Output:
- 1K, Pro 2K; WebP, PNG or SVG
- Access:
- Recraft (web), Recraft API
- Open weights:
- no
- Released:
- May 2026
xAI · United States
Grok Imagine Image 2.0
Generates images and edits several images together, for example to combine subjects or transfer a style.
- Inputs:
- Text, up to 5 images to edit
- Output:
- up to 10 images per request; formats up to 21:9
- Access:
- Grok apps, xAI API
- Open weights:
- no
Video
Models for video.
Create short clips from text or a start image, some with sound. Clip length is usually a matter of seconds.
Adobe · United States
Firefly Video Model
Video model, according to Adobe trained on licensed and public-domain content and designed for commercial use.
- Inputs:
- Text, optional first frame
- Output:
- resolution and length not stated for this model
- Access:
- Adobe Firefly (web)
- Open weights:
- no
Alibaba · China
HappyHorse 1.1
Text, first frame or references to video.
- Inputs:
- Text, first frame, reference images
- Output:
- 480p to 1080p; 3 to 15 seconds; with sound
- Access:
- Alibaba Cloud Model Studio (API)
- Open weights:
- no
Alibaba (Wan) · China
Wan 2.7 Video
Four models: text to video, image to video, reference to video and video editing.
- Inputs:
- Text, first and last frame, audio, video to continue, references
- Output:
- 720p or 1080p; 2 to 15 seconds; synced sound
- Access:
- Alibaba Cloud Model Studio (API), wan.video
- Open weights:
- no
- Released:
- April 2026
Alibaba (Wan) · China
Wan 2.6 Video
Text, image or references to video, with faster Flash variants.
- Inputs:
- Text, image, reference images and videos
- Output:
- 720p or 1080p; 2 to 15 seconds; synced sound
- Access:
- Alibaba Cloud Model Studio (API)
- Open weights:
- no
Alibaba (Wan) · China
Wan 3.0 Video
One model for generating, editing and extending, also from a document or web page; still a preview.
- Inputs:
- Text, first and last frame, up to 10 images, 5 videos, 5 audio clips, one document, one link
- Output:
- 480p to 1080p; 2 to 30 seconds; with sound
- Access:
- Alibaba Cloud Model Studio (API)
- Open weights:
- no
Black Forest Labs · Germany
FLUX 3 Video
Text, image or video to video with pinned keyframes and low-cost drafts first; still a preview.
- Inputs:
- Text, 1 to 10 keyframes, start video to continue
- Output:
- up to 4K, 24 fps; 5 to 20 seconds; with sound and speech
- Access:
- BFL API
- Open weights:
- no
- Released:
- July 2026
ByteDance · China
Seedance 2.0
Generates picture and sound together from four kinds of input; a Fast variant responds more quickly.
- Inputs:
- Text, up to 9 images, 3 video clips, 3 audio clips
- Output:
- 480p or 720p; 4 to 15 seconds; with sound
- Access:
- Dola, BytePlus (API)
- Open weights:
- no
- Released:
- February 2026
ByteDance · China
Seedance 2.5
Generates picture and sound together, understands reference videos, edits video and controls the camera.
- Inputs:
- Text, images, reference videos, audio
- Output:
- up to 30 seconds per run, extendable twice; with sound
- Access:
- Dola, Jimeng, BytePlus ModelArk (API)
- Open weights:
- no
Google · United States
Gemini Omni Flash
Generates and edits video conversationally, keeping characters and scene across turns; now recommended by Google as the default video model in the Gemini API.
- Inputs:
- Text, images (first and last frame, subjects), videos to edit and extend
- Output:
- up to 720p, upscaled to 1080p or 4K; extendable to 40 seconds; with sound
- Access:
- Gemini app, Flow, YouTube Shorts, Google Vids, Gemini API
- Open weights:
- no
- Released:
- June 2026
Google · United States
Veo 3.1 (Standard, Fast, Lite)
Video with sound, from text or between a first and last frame, with reference images and extension; still a preview in the Gemini API.
- Inputs:
- Text, first frame, first and last frame, up to 3 reference images (not Lite)
- Output:
- 720p, 1080p or 4K (not Lite); 4, 6 or 8 seconds, extendable to 148 seconds; with sound
- Access:
- Gemini app, Flow, Gemini API
- Open weights:
- no
- Released:
- October 2025
Kuaishou (Kling AI) · China
Kling 4.0 / 4.0 Flash
Long single takes, steered by several keyframes and references, lip-sync in 9 languages; initially in early access.
- Inputs:
- Text, up to 10 images, 5 videos, 10 keyframes, audio
- Output:
- up to 30 seconds; stereo sound
- Access:
- Kling AI (app); API not mentioned
- Open weights:
- no
- Released:
- September 2026
Kuaishou (Kling AI) · China
Kling Video 3.0
Several shots in one clip with sound and lip-sync, characters stay consistent.
- Inputs:
- Text, image, first and last frame, element reference
- Output:
- 3 to 15 seconds; with sound
- Access:
- Kling AI (web, app), Kling API
- Open weights:
- no
- Released:
- February 2026
Kuaishou (Kling AI) · China
Kling Video 3.0 Omni (O3)
Combines image, video and sound references in one model; voices of individual elements can be controlled.
- Inputs:
- Text, image, first and last frame, several image and video references
- Output:
- up to 15 seconds; with sound
- Access:
- Kling AI (web, app), Kling API
- Open weights:
- no
- Released:
- February 2026
Luma AI · United States
Luma Ray3.2
Frame-level control, continuity across cuts and HDR output for grading; does not generate sound.
- Inputs:
- Text, image, up to 16 keyframes, video to modify
- Output:
- up to 1080p, HDR and EXR; 5 or 10 seconds; no sound
- Access:
- Luma (web), Luma API
- Open weights:
- no
- Released:
- June 2026
MiniMax · China
MiniMax H3 (Hailuo)
One model for picture, voice, sound effects and music, handling mixed inputs in one prompt; the open weights deliver 768p, the 2K module is not open.
- Inputs:
- Text, images, video, audio
- Output:
- 2K (cheaper 768p); up to 15 seconds; stereo sound
- Access:
- Hailuo, MiniMax API; weights on Hugging Face (own community licence)
- Open weights:
- yes
- Released:
- July 2026
Runway · United States
Runway Gen-4.5
Runway’s text- and image-to-video model.
- Inputs:
- Text, image
- Output:
- resolution and length not stated
- Access:
- Runway (paid plans), Runway API
- Open weights:
- no
- Released:
- December 2025
xAI · United States
Grok Imagine Video 1.5 / 1.5 Lite
Video from image and voice references with pinned keyframes and lip-synced speech; Lite accepts text and image only.
- Inputs:
- Text, image, up to 14 reference images, 3 voice references, up to 4 keyframes
- Output:
- 1080p (Lite upscaled from 720p); 1 to 15 seconds; with sound
- Access:
- Grok apps, xAI API
- Open weights:
- no
- Released:
- July 2026
Missing a model or spotted an outdated detail? Let us know. Contact form
No longer listed because discontinued: OpenAI Sora 2 (24 September 2026), Google Imagen 4 (Gemini API, 17 August 2026) and the original Nano Banana (Gemini 2.5 Flash Image). Nano Banana 2 and GPT Image 2 have successors.
What to check
Before you use a model.
Rights and usage
Whether you may use results commercially depends on the terms of the provider you access the model through. For people, brands and third-party works, personality, trademark and copyright law apply as well.
Labelling
Since 2 August 2026, anyone publishing AI-generated or altered images, video or audio that look real (deepfakes) must disclose this under Art. 50 of the EU AI Act.
More on the EU AI Act →Original or reseller
Many sites carry well-known model names in their address but don’t belong to the maker; they resell access. Check who runs the service, where your data goes and which terms apply.
Data and location
Uploaded images and prompts go to the provider. For client material or images of people you need a legal basis and often a data processing agreement. Open-weight models can also run on your own hardware.
From the blog
An image model in real use.
Artificial Intelligence ·
AI Product Images: Why the Subscription Cost More Than Our Own Solution
Case study LANAKILA: how a platform subscription for AI product images became a custom desktop app connected directly to Google's image AI – and what that means per image.
Read the article →FAQ
Frequently asked questions.
Which model is best?
What does “open weights” mean?
Where do the details come from?
Do you help with choosing and integrating a model?
Is there a ready-made tool for AI product images?
Is a direct API connection worth it compared with a credit-based platform?
AI consulting
Which model fits your task?
We compare image and video models for your case by result, running cost and data protection, test them with your own material and show whether connecting directly to the maker’s API pays off – instead of a platform with subscription and credits.
From our work
RENEKI: AI image generation for teams.
For LANAKILA we built a desktop app that creates new visuals from product photos: with image models connected directly through the maker’s API, prompt building blocks at a click and the cost of every image. Designs and cutting patterns stay on the computer; only the finished image goes to the shop.
RENEKI is also available for your company: we set it up and, if you like, adapt it to your team, your models and your shop.
What it saves in cost →Looking for a ready-made tool rather than a model? Go to the AI tools directory →