Over the past year, image editing models crossed a quality line. Putting products into new scenes, swapping materials and colors, and producing campaign variants at volume became real production work.

Then came the catch: the best editing models also got slower and more expensive. Krea frames the trade precisely.

"A minute and a quarter of a dollar per edit is fine for one hero shot. For a whole catalog, or for trying ten ideas before lunch, it adds up fast."

Krea 3, announced on October 15, aims at the other side of that trade.

Krea 3 in one block

  Native 2K edits, mostly under 8 seconds end to end
  $0.06 per image at 2K — a price you can run on a whole catalog
  Up to 6 reference images per edit
  It leaves alone what you didn't ask it to touch
  (products, characters, and text stay intact)

The official comparison: 8s vs 33s vs 90s

These are Krea's published figures. Competitor timings come from actual edit jobs on Krea's platform.

Metric Krea 3 Nano Banana Pro GPT Image 2
Time per edit Under 8s 33s 90s
List price per image $0.06 (2K) $0.135 $0.22
Monthly cost, 100K edits $6,000 $13,500 $22,000

Footnote: competitor times are median request-to-finished-image latency for edit jobs on Krea, September 24 to October 1, 2026, queue included. Prices are list prices without image references; Krea 3 figures are at 2K. At 1K, Krea 3 costs $0.03 per image.

(The announcement also charts Krea 3 under 8 seconds against 27s–90s for other models. Either way, the story is single-digit seconds versus minutes.)

What matters more than the numbers is what they change. At eight seconds per edit, you stop writing careful one-shot prompts and start trying things: look, adjust, try again. The model stops being a step in the workflow and becomes the workflow.

What it preserves, what it changes

The hard part of editing is everything you did not ask for. Krea 3's feature list aims exactly there.

- Product integrity: keep the label and logo exactly as-is while
  building a new scene around it — a white-background packshot
  becomes a campaign photo
- Precision edits: "Replace only the seating in image 1 with the
  sofa, chairs, and table from images 2-4. Keep the rug, the room,
  and the reading woman."
- Sketch to render: a technical drawing plus a color palette becomes
  a photoreal render with the label kept 100% accurate (per Krea's examples)
- Character consistency: a few references put the same person or
  character on a new street, in a new pose, under new light
- Style retention: paintings, vector art, and ink drawings stay
  in their own medium

Background swaps, colorways, poster localization, ad-format adaptation, and material/light studies for buildings all sit on top of this combination.

Prompt patterns the official examples share

Krea's example prompts follow a common structure — worth copying directly.

1. Start with what stays
   "Keep the product exactly as it is" — name what matters,
   like the label or the logo

2. Describe the change the way a photographer would
   scene -> light -> camera. "Warm morning light, 35mm film look"

3. With several images, say what each one is for
   "Image 1 is the room, image 2 is the sofa"

4. Use "only" when one thing should change
   "Remove only the woman in the red coat"

Example: "Replace only the seating in image 1 with the sofa, chairs,
and table from images 2-4. Keep the rug, the room, and the reader."

This matches the point from our Muse Image piece on why editing now matters more than generation: the ability to steer an image safely through revisions beats the ability to generate one good image in a single shot.

What to read with caution

- The comparison figures are vendor measurements. Competitor timings
  were taken on Krea's platform, which may differ from those models'
  own best setups. Treat them as a relative signal (roughly 4x),
  not as absolute truth, until independent runs confirm them.
- "Mostly" under 8 seconds. Krea itself says the model is not
  real time yet, with the fastest setup landing at 4-5 seconds.
- Quality claims (label fidelity, character identity) rest on Krea's
  own examples. Verify with your product photos.
- List prices exclude image-reference costs; check the billing
  policy before running volume jobs.
- Character consistency carries the usual ethics questions about
  unauthorized use of real people. Start from assets you have
  the rights to.

CodeBridge mini lab: stress-testing edit durability

The briefing's action, turned into an experiment:

1. Pick 3 test images (a product shot, a portrait, a poster with text)
2. Run two edit types on each, 10 times each:
   [ ] background change ("studio background -> outdoor scene")
   [ ] object removal ("remove only the woman in the red coat")
3. Record:
   [ ] edit success rate (did only the requested thing change?)
   [ ] text and logo preservation (poster test)
   [ ] median time per edit
   [ ] real cost per image, reference images included
4. Decide by: do "8 seconds and $0.06" hold on your material?

Conclusion: the axis of competition moved

Yesterday's question: "Which model produces the best edit?"
Today's question: "How cheap and fast is one revision cycle?"

Krea 3's real claim is throughput, not peak quality. It targets the point where 100,000 catalog edits stay affordable — where trying more ideas beats polishing one attempt. Read it as a signal: the way we choose image tools is moving from best-single-image quality to speed times cost per revision.

Further reading

References

Go deeper with a course

Knowing one editing model matters less than sensing how images move through a content workflow — text, images, video, revision after revision. Start with the fundamentals of making things with AI tools.