A stock library and an AI suite on one subscription, with a node canvas wiring the tools together.
Freepik pairs a large library of photos, vectors and templates with an AI suite, now carrying the Magnific name, that spans image, video and audio models. Spaces is its node canvas, where tools connect into a pipeline and teammates work with shared cursors.
Best for: Designers who also need stock
- The stock library covers photos, vectors and templates, so not every asset starts from nothing.
- A design editor, a mockup generator and an icon generator sit next to the models.
- Spaces is a node canvas, so a repeatable setup means wiring tools by hand.
- Video runs through a generator, a clip editor and a separate project editor.
The aesthetic benchmark for AI imagery, video included, sold only on paid subscriptions.
Midjourney sells one product, its own model family, currently V8.2, through a web app and through Discord. Video generation is live with SD and HD output, and concurrency is the shape of the product: three image prompts at once on the two lower tiers, twelve on the two upper ones.
Best for: Art direction and concept imagery
- The model line is bought for its look, which is what its users say they stay for.
- Style and character references hold one look across a whole series of images.
- There is no free tier, so evaluating the model starts on a paid plan.
- Below Pro, generations are visible to the community, and companies past a million dollars in annual revenue sit on Pro or Mega.
The open image model built around typography, with Ideogram also on the Morphic image roster.
Ideogram 4.0 ships as a web app with an API and an MCP surface beside it. Text inside the picture built its reputation, and it is still the reason a poster or a packaging comp goes here first. The free plan issues ten slow credits a week.
Best for: Posters, packaging and set type
- Every plan clears its output for commercial work, the free weekly credits included.
- API and MCP surfaces make the model reachable from software you already run.
- The product is image only; the moving, sounding half of a campaign happens elsewhere.
- It lives outside the Adobe stack, so comps re-enter Photoshop by export.
#5
DALL·E (now ChatGPT Images)
OpenAI is retiring the DALL·E GPT and pointing image users at ChatGPT Images instead.
OpenAI dated the change itself. A release note on 31 July 2026 gives 30 August 2026 as the retirement date for the official DALL·E GPT, and the dall-e-2 and dall-e-3 API models had already gone on 12 May 2026. The successor family is GPT Image, currently gpt-image-2, which reaches 4K in a single generation.
Best for: Image work inside a chat assistant
- The API is documented and supports Batch, a route Morphic has no equivalent for.
- Free-tier images carry no visible watermark, and ownership of what you generate is yours.
- The DALL·E name retires on 30 August 2026, with ChatGPT Images as the successor.
- Consumer tiers default to your content being used for training, with an opt-out; Business reverses that.
Image and video generation with mask-based refinement and a canvas that draws as you sketch.
Leonardo AI generates images and video and wraps real editing tools around them. Canvas Editor refines a picture through masks, the closest thing on this list to working a Generative Fill selection. Realtime Canvas previews the image while you sketch, so composition becomes a drawing decision.
Best for: Sketching and mask-based refinement
- Mask-based refinement handles the fix-this-part-only edit that design work keeps asking for.
- Realtime sketching settles composition before the prompt gets precious.
- The strengths sit on the image side, around refining a frame rather than assembling a sequence.
- Canvas Editor, Realtime Canvas and Flow State are separate tools rather than one surface.
Real-time multi-model canvas with strong finishing tools, and Krea models are on Morphic too.
Krea puts its own models next to third-party image models on one canvas, and the signature behaviour is real-time: the picture updates while you type or draw. Upscaling and enhancement are the finishing tools people stay for.
Best for: Fast iteration and finishing
- Real-time generation turns iteration into something continuous rather than a series of submissions.
- Team tiers cover shared assets, role-scoped permissions and billing for more than one seat.
- Krea Nodes builds a repeatable process by connecting nodes, a different habit from describing one.
- Strong at making and polishing frames; assembling them into a cut is not the product.
Open weights you download and run yourself, with no per-generation meter once the hardware is yours.
Stable Diffusion 3.5 is the flagship image family from Stability AI, and the Core Models list carried no image successor when it was last revised in May 2026. The Community License allows commercial use at no charge under a million dollars in total annual revenue, carries a mandatory on-screen credit reading "Powered by Stability AI", and can be revoked if you break its terms.
Best for: Running the model on your own hardware
- Once the weights are local, a generation costs hardware time instead of a credit balance.
- Fine-tunes and LoRAs are permitted, and selling the derivatives you train is permitted too.
- No ongoing free hosted tier remains: the API opens with 25 free credits, then bills per generation.
- Nothing newer than SD 3.5 has appeared on the open image side since October 2024.
