Published January 27, 2025 · By CampaignsLive · Insights
The resolution ceiling on commercial AI image tools moved more in 2024 than in any prior year. Tools that had been capped at two to four megapixels in 2023 ended 2024 producing twelve to sixteen megapixel output at acceptable quality. The change unlocked categories of brand creative production that had not been available before — full-bleed print, billboards, magazine spreads, environmental placements — and shifted the working assumption about where AI imagery could and could not be used.
This is a working timeline of when each part of that shift actually landed, and what it unlocked at each step.
The 2023 ceiling
At the start of 2024, the working state of the art looked like this.
Midjourney’s v5.2 release produced output at roughly 2-megapixel resolution. The model’s upscaler produced higher-resolution variants that were acceptable for medium-format use (small-format print, A5 print, large-format digital) but began to show interpolation artifacts at A4 and above. Producing usable 12-megapixel imagery from Midjourney in 2023 required a stack of tools downstream of the generation: external upscalers like Topaz Gigapixel or Magnific, careful detail-recovery work, often a touch-up retouch cycle for hands and faces. The pipeline was workable but slow.
Stable Diffusion’s SDXL release similarly capped at around 4 megapixels for base generation. The open-source upscaler ecosystem was more flexible than Midjourney’s, but the same fundamental trade-off applied: higher-resolution output required additional steps and additional artifact management.
DALL·E v3 produced output capped at roughly 1024×1024 — less than 1 megapixel — with no native upscaling. Image generation in DALL·E was effectively concept-grade only.
The aggregate effect at the start of 2024 was that AI imagery could be produced at print-resolution, but the production pipeline required either external tools or significant manual cleanup. Most brand teams that were producing print-grade AI imagery were running this stacked workflow, which was expensive enough in time and operational overhead that it limited adoption.
The 2024 milestones
Four shifts through 2024 moved the resolution ceiling.
Q1 2024: Higher-resolution diffusion checkpoints. The open-source community produced SDXL variants and follow-on releases that generated natively at 8 to 12 megapixels with acceptable quality. The change required no new model architecture; it was a function of training the existing architectures on higher-resolution data and tuning the inference parameters. By the end of Q1, working teams could produce 12-megapixel output from Stable Diffusion variants without the external upscaling step.
Q2 2024: Midjourney v6 with native upscaling. Midjourney’s v6 release moved the base-generation resolution upward and integrated an upscaler that produced acceptable 16-megapixel output as a single-pass workflow. The improvement was less in raw quality — the v6 outputs were not dramatically better than v5.2 upscaled through Magnific — and more in pipeline simplicity. The two-step workflow became a one-step workflow.
Q3 2024: Specialized print-resolution tools. Several smaller, more focused tools emerged that targeted the print and OOH categories specifically. These tools made resolution and color management their core proposition, in distinction to the general-purpose image tools that were optimizing for visual quality at the typical digital-display sizes. The market for tools whose explicit positioning was “print-grade” stabilized around several offerings, of which CampaignsLive was one.
Q4 2024: Fine-tuning workflows for print. The deeper change of 2024 was that fine-tuning workflows — where a model is adapted to a specific brand’s visual style on a specific corpus — became operationally tractable. This had been technically possible since 2023 but had been slow, expensive, and required engineering investment most brands did not have. By Q4 2024, the tooling around fine-tuning had matured enough that brand teams could produce custom-trained models at production quality without engineering investment.
What this unlocked
By the start of 2025, the resolution ceiling on commercial AI image tools had reached a place where the following categories of brand creative production were operationally feasible:
- A4 magazine page at 300 PPI. Native 12-megapixel generation produced full-page magazine work without upscaling.
- Double-page magazine spread. 16-megapixel output handled spreads with bleed and crop allowance.
- 6-sheet posters at 1.2m × 1.8m. The smaller OOH formats were comfortably inside the new resolution envelope.
- 48-sheet roadside billboards. Print-spec PPI for OOH (lower than press print, because viewing distance is greater) brought 48-sheet inside the envelope as well.
- Bus shelter transit panels. Standard transit formats at full bleed.
- Retail point-of-sale. Window vinyls, end caps, shelf wobblers, hangtags — all comfortable.
- Standard editorial cover work. 16MP handled editorial covers without the close-detail failures that had plagued 2023’s output.
The categories still beyond the envelope, at the start of 2025, were the largest formats: wallscapes, building wraps, airport dominations, environmental work at scale. These required either native generation above 30 megapixels (which the commercial tools mostly did not offer) or a generation-plus-upscaling workflow that the smaller brand creative teams could handle but the largest could not yet do at production volume.
What did not change at the same pace
Three things that the resolution ceiling did not fix.
Color management. Native generation in 2024 remained mostly sRGB. CMYK output, ICC-profile-correct production for specific printers, and the broader print-prep workflow stayed the same problem the 2023 tools had. The tools that solved it were the specialized print-focused ones; the general-purpose tools did not. For a brand team running a hero asset across digital and print, the color workflow was still the friction point.
Identity across compositions. Higher resolution did not solve the question of generating the same character, the same product, or the same environment across multiple compositions. The fine-tuning workflows that matured through 2024 helped with this — a fine-tuned model on a specific subject could produce consistent identity across compositions — but the base-model identity-consistency problem was not solved by resolution improvements.
Reliable text in images. Per the 2023 review, generating legible text within an image remained unreliable through 2024. The brands that needed in-image typography continued to handle it through traditional design overlay on AI-generated base imagery, which was an acceptable workflow but not a solved one.
Where this leaves things
The resolution shift through 2024 effectively brought AI imagery from “concept-grade for digital” to “production-grade for digital, print, and most OOH.” The categories of brand creative work that were not operationally feasible with AI tooling in late 2023 became feasible by the end of 2024.
The bottleneck moved from raw output capability to the surrounding production stack — color management, identity consistency, brand-asset integration, downstream production pipeline. The tools that have done well in 2025 have been the ones that focused on these production-stack problems rather than continuing to push raw output capability.
For the production-prep side of the resolution question, see Generating Print-Resolution AI Images. For the specific use case of OOH at full resolution, see the Print and OOH page.