Make Things: Images, Video, and Audio in 2026
This content was developed with AI assistance and is regularly reviewed for accuracy.
Snapshot: October 2026. Model names, prices, and product features change monthly; What's New in AI is the maintained chronology.
Two years ago, making an AI image meant learning a separate app and its parameter syntax. In 2026 the default path is to ask the assistant you already use, then refine the result in conversation. Specialist tools still lead for cinematic video, high-end aesthetics, and music, and the field changes fast enough that OpenAI's much-hyped video app, Sora, launched and shut down within six months. This page gives you a working method that survives those changes, the current starting points, and the rights questions to settle before you publish.
Learning focus
- Create and edit images inside ChatGPT or Gemini with a short visual brief
- Know when to move to a specialist tool for video, voice, or music, and which ones lead as of October 2026
- Edit generated media in small steps and check it for the errors these tools still make
- Handle licensing, consent, and disclosure before anything goes public
Start in the chat app
ChatGPT and Gemini both generate and edit images inside the conversation, and both offer it on their free tiers with usage limits: ChatGPT through OpenAI's GPT Image models and Gemini through Google's Nano Banana family. Microsoft's Copilot has its own in-house image model. Claude is the exception; it does not generate images, audio, or video, though it is useful for writing the brief, critiquing drafts, and building diagrams or slides as files.
The conversation is the feature. You can attach a photo and say "remove the person on the left and make it dusk," then "now a square version for Instagram," then "put our logo bottom right, small." Each step builds on the last, which is far easier than rewriting a long prompt from scratch.
The brief from the Brief lesson still applies, with visual fields:
Purpose: [where this image will be used and what it must communicate]
Subject: [what is in it, specifically]
Style: [photo, illustration, flat vector, watercolor; or "match the attached reference"]
Composition: [angle, framing, what goes where, empty space for text]
Text in image: [exact wording, or none]
Format: [aspect ratio and size, e.g. 16:9 banner, 1:1 post]
Avoid: [anything that must not appear, e.g. no people, no brand logos]
Reference images beat adjectives. If you want a house style, attach two examples you like and say "same palette and lighting." If you need a character or product to stay consistent across several images, attach the first approved image each time and say what must not change.
When to use a specialist tool
Move beyond the chat app when you need something it does not do well: longer or more cinematic video, a particular aesthetic, synthetic voice, or full songs. The tools below are the ones 2026 reviewers cite most often. Rankings are drawn from secondary reviews and shift every few months, so treat this as a starting list, not a verdict; current plans and prices live on the AI Tools & Platforms and Free AI Tools Directory pages.
| You want to make | Start with | Also consider |
|---|---|---|
| Images for everyday use | The image feature in ChatGPT or Gemini | Canva's AI tools for layouts and templates |
| Stylized, high-aesthetic images | Midjourney (paid only, no free tier) | Open-weight FLUX models if you want to self-host |
| Edits to real photos | Adobe Firefly inside Photoshop | The chat apps for quick fixes |
| Video clips with sound | Google's Veo 3.1, through Gemini or Google's Flow | Kling 3.0 and ByteDance's Seedance 2.0; Runway for fine control |
| Presenter or training video | HeyGen or Synthesia avatars | CapCut for editing what you shot |
| Voiceover and narration | ElevenLabs | The voice modes in the chat apps for drafts |
| Music | Suno (v5.5 since March 2026; a v6 release was reported in September) | ElevenLabs' music tools |
| Cleaner recordings and transcripts | Descript or Adobe Podcast | Meeting notetakers such as Granola for calls |
Two 2026 events are worth knowing because they teach the same lesson. OpenAI announced Sora's shutdown in March 2026, took the app offline in April, and removed the video API in September, with no successor named. Udio, a leading music generator, disabled downloads after settling with major labels and had not completed its promised licensed relaunch by late September. Tools disappear and change terms. Keep your own source files, references, and briefs outside any single app so you can move.
Edit in steps, then check
Generated media fails in predictable places, so build a short review into every job.
Change one thing at a time. "Make the sky warmer" followed by "move the headline up" gives you control; a paragraph of changes produces a new image with new problems. When an edit goes wrong twice, go back to the last good version and try a different instruction, the visual version of the two-correction rule in Check.
Work from stills to motion. For video, approve a storyboard of still frames first, then animate the frames you like with an image-to-video tool. It is cheaper, faster, and more predictable than regenerating whole clips.
Check the usual suspects before anything ships:
- Text in the image: spelling, legibility, and whether it says what you asked
- Hands, faces, and bodies: extra fingers, mismatched eyes, impossible poses
- Products and logos: accurate shapes and colors, nothing that resembles another company's mark
- Scale, lighting, and shadows that agree with each other
- Audio sync, mispronounced names, and voices that drift between takes
Review for representation. Image models default to stereotypes about who appears in which role. Say who should appear when it matters, and look at a batch of results together to spot the pattern.
For designers and creative professionals
Professionals get the most from these tools at the edges of the work rather than the center: exploration before the concept, production after it. Useful patterns:
Generate [number] distinct visual directions for a [project type] for [audience], each with a one-line rationale, a palette, and a sample image. Vary the style widely.
Create a product mockup of [product] in [setting], matching the attached brand guide's colors and type. Leave the top third empty for a headline.
Put your brand guide, approved references, and style rules in a project so every session starts from them, and turn a reliable workflow into a skill as described in Supply Context. Keep final production in the tools that give you layers and precise control; treat generated images as concepts, mood boards, backgrounds, and variations to test. Tell clients where AI was used, and agree in the contract who owns what.
Rights, consent, and disclosure
Settle three questions before anything you generate goes public.
Can you use it commercially? Usage rights depend on the tool and your plan. Read the terms of the tier you are on, and prefer tools that state commercial use plainly for client work. Copyright law around AI is moving: the first major AI copyright settlement, $1.5 billion in Bartz v. Anthropic over pirated books used in training, received final approval in July 2026, and other suits over images and music are still pending. Avoid prompts that ask for a named living artist's style or a recognizable character for commercial use.
Do you have consent? Never clone a real person's voice or generate their likeness without their permission, and never create realistic media that could pass as a real event. Upload photos of other people only with their agreement; the Protect lesson covers why.
Have you disclosed it? Follow the platform's labeling rules and your client's or employer's policy, and lean toward a simple label such as "image generated with AI." Regulation is heading the same way: the EU AI Act requires machine-readable marking of AI-generated content, with that obligation reported to apply from February 2027 for systems already on the market.
Key takeaways
- Start in ChatGPT or Gemini, brief the image like any other task, and edit in conversation; Claude does not generate media
- Move to a specialist tool for cinematic video, high-end aesthetics, voice, or music, and expect the leaders to change
- Change one thing at a time, go from stills to motion, and check text, hands, logos, and sync before shipping
- Tools disappear: keep your references, briefs, and source files outside any single app
- Confirm commercial rights, get consent for any real person's voice or likeness, and disclose AI-generated media
From the author
Designers can also use agents for usability testing. Instead of writing a test script and recruiting participants, you can have an AI agent play the first-time visitor. FreshVisit, built by the author of this site, takes a task like "find how to cancel," tries to finish it on desktop, tablet, and phone, then scores how findable it was and flags accessibility issues on every page it reached.
Next steps
Continue to: Make Things: Build an App Without Code
- Compare current plans and free tiers in the Free AI Tools Directory
- Find ready-made visual and creative briefs in the Brief and Prompt Library