Generative AI for designers: stop betting your workflow on a model (Sora has just died)

Rankings change every quarter - Sora has just paid the price!

If you want proof that no one should build their creative workflow around a specific AI tool, take a look at Sora. OpenAI's flagship video generator, heralded just over a year ago as the future of creation, is shutting down: the web and app experience ends on April 26, 2026, the API in September. The reason is brutally simple - around $15 million in costs per day for $2.1 million in cumulative revenues over its lifetime. The industry's most hyped product didn't last two years.

I tell this because it's the central lesson of this field, and the one that most "top 10 AI tools" articles forget to draw: the names I'm about to mention will be partially wrong by the time you read this. Midjourney went from v6 to V7 to V8 Alpha in less than a year; DALL-E was replaced; Sora disappeared. What hasn't changed is a professional designer's method of integrating these tools without getting trapped. Let's start with the card of the moment, then I'll move on to what lasts.

The map of image generators in 2026

Here's the actual state as I write, keeping in mind that each line can move from one quarter to the next.

Midjourney V8 Alpha, launched on March 17, 2026, generates around five times faster than V7 - what used to take 30 to 60 seconds now takes less than 10 - with significantly better text rendering and native 2K via --hd. But to integrate legible text into a visual, I use GPT Image 2 or Ideogram V3, which are more reliable in this respect. And as far as open source is concerned, FLUX has overtaken Stable Diffusion in popularity: its pro, dev and schnell variants run locally, enabling a template to be trained on a brand's assets and guaranteeing a stylistic consistency that no proprietary tool gives.

For customer work, the criterion is not beauty

This is where many designers go wrong: for a billed deliverable, the most beautiful model isn't the right one - the right one is the one that doesn't expose you legally. Adobe Firefly Image Model 5, released in October 2025, is the only large model trained exclusively on licensed and public domain content, eliminating the intellectual property risk that hangs over models trained by massive scraping. It's my default choice for retouching and extensions, supplemented by FLUX.1 schnell when I need permissively licensed open source.

The question of copyright is not theoretical. The US Copyright Office, in the second part of its report published on January 29, 2025, confirmed that an AI-generated image is only protectable if a human has made a sufficient expressive contribution to it - simply writing prompts is not enough. The D.C. Circuit Court of Appeals confirmed this line in the Thaler case, and the Supreme Court declined to hear the case on March 2, 2026: the requirement of a human author is now firmly established in the United States.

Generative video has moved to 2026

The video market has completely reconfigured itself. With the departure of Sora, three players share the field. Google Veo 3.1 offers the best overall quality: native 4K, and above all synchronized audio - ambience, dialogue, effects - generated in a single pass with the video. Runway, with Gen-4 and Gen-4.5, remains the favorite of pros who want granular control: camera movements, motion brush, character consistency by reference. Kling 3.0 completes the top three. My reflex for a customer deliverable requiring precision is Runway; for a piece that needs to show off with integrated sound, Veo 3.1. The Sora lesson applies here too: these are tools of the moment, not foundations.

My method doesn't depend on the model

Here's what I do, and what hasn't changed in three years, despite the many versions. Generative AI lives in my ideation and exploration phase, never in the final deliverable. Instead of spending hours on image banks looking for the right atmosphere, I quickly generate ambient visuals that crystallize an artistic direction - then finalize in Photoshop or Illustrator. AI results are sketches, not finished products.

For client work, three firm rules: I use commercially safe tools, I systematically document prompts and retouching steps for traceability, and I ensure significant human creative input on every deliverable - both for quality and copyright purposes. And I always disclose the use of AI to my clients. Generative AI is an extraordinary creativity amplifier, and a great teaching tool for showing a customer what a direction would look like in a few minutes. But it doesn't replace the designer's vision, critical sense or responsibility. Models will come and go; this method will last for the next version cycle - and the one after that.