Skip to content
AI Primer
release

Midjourney V8 Alpha adds faster 2K renders as tests flag softness and anatomy errors

Creator tests say V8 is faster and text output is better, but many posts report softer images, profile bleed, mixed prompt adherence, anatomy errors, and heavier moderation. Run V7 versus V8 checks on your core prompts before switching full workflows.

3 min read
Midjourney V8 Alpha adds faster 2K renders as tests flag softness and anatomy errors
Midjourney V8 Alpha adds faster 2K renders as tests flag softness and anatomy errors

TL;DR

  • Midjourney V8 Alpha's first wave of creator tests says the model is fast and visually clean, with several posters calling out stronger aesthetics even when they dislike other changes, according to first V8 tests, robot tests, and follow-up.
  • Multiple users also say V8 is coming out softer than expected: softness complaints described missing texture and sharpness, while side-by-sides from anime set compare and cinematic compare argue V7 still wins on detail and anatomy.
  • Text handling looks improved in at least some creative use cases: poster thread says text rendering is better, and comic split shows a cleaner split-poster composition than the creator had been getting before.
  • The bigger issue is consistency. Early posts report misgendering, heavy profile bleed, proportion errors, weak prompt adherence in watercolor sketches, and stricter moderation, as seen in first impressions, watercolor tests, and poster thread.

What actually improved

The clearest upgrade in day-two testing is speed. Early testers repeatedly describe V8 as fast, and several pair that with cleaner-looking outputs rather than a dramatic aesthetic reset. _VVSVS called the first renders "quite clean" and later said the model already shows "great potential and clean output," while 0xInk said V8 is "incredibly fast" and promising in robot-fashion portraits first impressions robot tests follow-up.

Text is also landing better in at least some poster workflows. Artedeingenio wrote that text rendering has improved, then posted eight poster results with readable titles built from prompts in the image alts; the same creator later shared a comic-style Hulk/Thor split image as a "big improvement" case for V8 comic split. That does not mean prompt following is fixed across the board, but it is one concrete area where creators are seeing gains.

Where creators are seeing regressions

Softness is the most repeated complaint. Julie W Design said many V8 images "lack texture and sharpness," and a later V7-versus-V8 portrait test found that turning off --p and adding --exp did not help much; Midjourney's own account replied by asking whether the issue also appeared on plain prompts and whether --hd had been tried v7 v8 portrait Midjourney reply.

Several creators are also flagging weaker control. _VVSVS reported misgendering on a basic "a king" prompt, heavy profile influence, human proportion issues, and recognizable franchise characters appearing in outputs first impressions. Artedeingenio said V8's watercolor sketches changed significantly from V7 and did not strictly follow prompts for handwritten notes watercolor tests. A Turkish tester described persistent anatomical errors, objects merging into each other, and overly plastic textures in repeated V7 prompts, while ProperPrompter's anime talk-show comparison also favored V7 on the same prompt and references cinematic compare.

The early read from side-by-side tests

The broad pattern is not "V8 is bad" so much as "V8 is uneven." Techhalla framed the release as "a choice" after running seven prompts against Midjourney and a larger 13-prompt comparison in Nano Banana 2 on Freepik using unlimited mode 13 prompt note. Separate quick plays still show V8 producing stylish motion, fashion, and automotive imagery, which helps explain why even disappointed users keep calling Midjourney the "king of aesthetics" quick play aesthetics take.

That split verdict is the story of the alpha so far: faster, often attractive, occasionally better at text, but still unreliable on anatomy, fidelity, and exact prompt control in the workflows creators care about most.

Further reading

Discussion across the web

Where this story is being discussed, in original context.

On X· 4 threads
TL;DR5 posts
What actually improved4 posts
Where creators are seeing regressions5 posts
The early read from side-by-side tests3 posts
Share on X