I'm trying to build a case for my team to get a budget for AI image generation, specifically DALL-E 3 via the API. The obvious pushback is: "We already pay for a stock photo subscription. Why do we need this?"
On paper, $50/month for "unlimited" downloads from a major stock site seems cheaper than generating images via OpenAI's API. But I'm not convinced the comparison is that simple.
With stock, I spend a lot of time searching for the *exact* concept, often settling for something "close enough" that doesn't perfectly match our product's context. There's also licensing headaches for niche use cases.
With DALL-E 3, I can generate a hyper-specific image of, say, "a dashboard graph showing a 30% week-over-week increase in user retention, in our app's exact color scheme, with a minimalist aesthetic." That's powerful for blog posts and internal presentations.
My back-of-the-napkin math: if I need 20 unique, highly-tailored images a month, and each is a standard 1024x1024 generation ($0.04), that's $0.80. Even at 100 images, it's just $4. The cost seems to be in the time spent crafting prompts and doing revisions.
But I'm new to this. Has anyone actually run the numbers for regular production use? Where does the real cost/benefit tipping point lie? Is it in the iteration time, or are there hidden API costs I'm missing? 🤔
I'm a solo admin at a 40-person B2B SaaS company, handling our marketing tools and user onboarding. I've used both stock subscriptions and DALL-E 3 for generating blog and help center images.
1. **Real monthly cost**: Stock is a fixed $50, but DALL-E 3 is variable. Your $4 estimate is low. In practice, I spend $8-15/month because I do 2-3 revisions per final image and sometimes generate multiple options. The real cost is prompt engineering time, which can add 10-15 minutes per image.
2. **Licensing and legal risk**: Stock is simple for standard commercial use. DALL-E 3's terms grant you ownership, but you're responsible for the output. We had our legal team review it, and they flagged potential trademark or likeness issues in generated images as a gray area we now manually check.
3. **Output consistency**: DALL-E 3 struggles with repeatable characters or exact branding. You can't guarantee the hex code for your brand blue will be identical across 20 images. Stock libraries have consistent, professional styling, but you sacrifice specificity.
4. **Workflow integration**: Our stock site plugs into Canva and our CMS. DALL-E 3 requires generating in OpenAI's playground or via API, downloading, and then uploading to our asset manager. It adds two extra steps that break the flow for our designers.
I'd recommend DALL-E 3 only for highly specific, one-off images for internal decks or concept mockups where perfect branding isn't critical. For all public-facing blog and web images, we stick to the stock subscription. To decide, tell us the percentage of images that need your exact product UI and whether you have designer time allocated for prompt work.
You're spot on about the variable costs and legal gray areas. Where I'd add a caveat is on the time investment for prompt engineering. That 10-15 minutes per image is a learning curve cost. Once you develop a library of effective, company-specific prompts, the generation time per asset drops dramatically. I've templatized prompts for things like "hero image for a blog post about [topic] in [style]" that we reuse, cutting that time down to under 5 minutes for routine work.
Your point on output consistency is the real operational bottleneck, though. For any project requiring a coherent visual story - like a sequence of tutorial images with the same UI - the lack of deterministic output becomes a major time sink. You end up generating dozens of variants and manually stitching together the usable elements, which can negate the time saved from not searching a stock site.
Mike
Your math on generation costs feels right, but the time cost is bigger than you think. The first few months, that "10-15 minutes per image" from the other post is real. You're paying for that learning curve with your salary.
Have you factored in the extra software? You'll need an image editor for basic crops and fixes on the DALL-E outputs. That's another subscription or tool to manage.
The licensing point is huge for B2B. Our legal team would have the same concern. Who checks every generated image for potential trademark issues? That's a new manual step stock sites handle for you.
Agree on point 3. That consistency issue is a production killer. You can't build a visual system on "close enough" branding colors. I've had to enforce style guides with pixel-level accuracy in Grafana dashboards for executive reports - stock gets you 90% there, DALL-E 3 gets you a different 90% every time.
Your $8-15/month is optimistic for any real volume. It scales linearly, which a subscription doesn't. I track costs like an SLA. My benchmark: if you generate more than 50 final images a month, DALL-E 3 API costs will blow past that $50 stock fee, ignoring the labor.
Metrics don't lie.