REVIEWS / AI IMAGE / OWNER INSIGHTS
🦉 WE READ 332 OWNER COMMENTS
Midjourney: what owners actually say
Owners find Midjourney impressive and addictive but wrestle with reproducibility, control, and ethical unease about how training data is sourced
What owners complain about
- Difficulty controlling output SOME
Users report that getting exactly the image you want is hard — 'AI image generation just isn't quite there yet' and 'it still requires skill to get what you want.' You can never be sure a posted image came out perfectly on the first try.
- Reproducibility is inconsistent SOME
Even with seeds and fixed parameters, results vary. On Stable Diffusion, batch generation with all parameters fixed produces slightly different images, and results differ across operating systems and hardware (CPU vs CUDA vs MPS).
- Discord-only interface FEW
Users question whether they are 'restricted to Discord for usage' and ask if an API endpoint exists, suggesting the Discord-only workflow is a friction point.
- Anatomy and quality degradation from AI-on-AI training SOME
Commenters warn that training on AI-generated images creates a feedback loop — 'like when a mic picks up its own signal' — producing disfigured anatomy and increasingly 'samey' outputs, a model collapse concern.
- Copyright and fair use backlash COMMON
Users feel companies ingested collective creative work without compensation and question whether 'fair use' applies when the scale is industrial. Strong sentiment that the tech is 'plundering' a commons rather than building on it.
What owners love
- Addictive, fun image creation
Multiple users describe being unable to stop generating images — 'can't stop generating images ... it's so amazing' — and call it 'good fun' and 'pretty incredible.'
- Rapidly improving quality
Users note AI image generation is 'getting incredible' and praise the speed of improvement, with excitement about creating films, comics, and quick content from prompts.
- Creative workflow potential
Owners see real utility for fast content creation — people are 'already making comics with it' and envision it as a future content-creation pipeline.
- Seed-based reproducibility
Technical users appreciate that setting a seed gives some control over output reproducibility, enabling iterative refinement of a specific composition.
Surprising patterns
- Reproducibility is hardware-dependent: identical seeds and parameters yield different results on different OS, CPU, or GPU backends — a non-obvious gotcha for anyone doing serious work.
- Some users argue the human selection process behind submitted AI images means training data isn't purely random garbage — there's a curation step — which complicates the 'model collapse' narrative.
- A strong undercurrent of users who use the tool still express moral discomfort about it, calling the technology itself potentially beneficial while condemning its use 'to steal art and cut people out of the picture.'
WHO SHOULD SKIP IT
Anyone who needs pixel-level control, deterministic reproducibility across hardware, or a workflow outside Discord, and anyone for whom the copyright and ethical dimensions of AI-trained-on-human-art are a dealbreaker.
Synthesised from 332 real owner comments across 5 platforms. Every point is grounded in the comments — no marketing, no AI guessing. How we do it →