Regenerate with a different provider, just for one post
Sometimes one post's image is just wrong. The per-post engine picker lets you swap to a different provider for that post without changing the campaign default. Here's when and why.
You ran a 14-post campaign. AI generated 14 captions and 14 images. Twelve of them are great. Two are off — wrong era, wrong mood, wrong aesthetic. You don't want to regenerate the whole campaign. You don't even want to regenerate those two with the same provider that produced the misses.
That's the per-post engine picker.
What it is
In the wizard's Media studio, every post has a "Try a different engine for this regenerate" link next to the prompt. Click it and you get a picker listing the AI media providers — and only the ones you actually hold a key for are selectable (Pollinations images work with no key at all). Pick a different one, hit Regenerate, and the image (or video) for that post regenerates using the engine you picked.
The campaign's default provider doesn't change. Every other post keeps using the default. Just this one post swaps — and once it lands, the post shows a "Made with provider · model" line so you always know which engine produced what.
Why per-post, not per-campaign
Different providers have different strengths. Magnific is strong on cinematic photoreal. fal.ai is strong on stylized illustration because you can pick a specific model. Gemini Nano Banana is strong on Google-aesthetic clean editorial.
For a 14-post campaign your default is whatever fits the overall vibe. But specific posts have specific needs:
- Post 3 is a moody coastal scene → Magnific Mystic
- Post 7 is a watercolor sketch of a coffee shop → fal.ai with a watercolor model
- Post 10 is an abstract editorial background for a quote card → Gemini Nano Banana
- Every other post → your default (whatever it is)
The per-post swap gives you that flexibility without the overhead of running three campaigns or rerolling the whole thing.
When to use it
1. The image is wrong. Wrong era, wrong character implication, wrong mood. First try regenerate (same provider, new attempt). If you get the same flavor of wrong twice, swap providers. Different models have different "wrongs."
2. The image is generic. AI image models default to inoffensive-but-bland when the prompt is weak. If your default provider keeps producing generic, swap to fal.ai with a more specific model — sometimes the model picker is what unlocks specificity.
3. The image fails entirely. Rate-limited, API timeout, key expired. The post shows the specific error. Swap providers for just that post instead of fixing the failed provider's account in the moment.
4. Aesthetic mismatch within a campaign. Your campaign is mostly photoreal but one post calls for an illustration moment (a quote card, an explainer panel). Swap to fal.ai with an illustration model just for that post.
5. You're testing. First time using a new provider. Run a campaign with your usual default, then regenerate a few posts on the new engine to see how it compares against the same prompts. Cheaper than running parallel campaigns.
When NOT to use it
1. You haven't tried regenerate yet. Same provider, new attempt is the first move. Different attempts on the same prompt often produce different results. Save the provider swap for when you've ruled out variance.
2. The whole campaign is wrong. If 10 of 14 posts are missing the mark, the issue isn't the provider — it's the brief. Fix the objective or your persona's Image Style Guide (editable right in the wizard's Brand Look card) and regenerate the batch.
3. Visual cohesion matters more than the specific image. A launch campaign where every post needs to look like part of a deliberate set probably shouldn't have one outlier in a totally different aesthetic. Stick with one provider for that campaign.
Skipping a post
Adjacent to the regenerate controls is Skip. Sometimes the right answer isn't a different image — it's no post at all. Skip a post and it won't be generated or scheduled. The campaign continues.
Use cases:
- A concept that doesn't land for you and isn't worth rewriting
- A slot you'd rather fill manually with something real-time
- A post where AI keeps failing and you'd rather skip than fight it
The batch still ships the other posts. Skipping doesn't break the rhythm; it just removes one post from the queue.
Upload-Your-Own
Batches created with the "upload my own media" source swap the generate palette for an upload slot on every post. Drop in your own image or video; the caption stays.
Use cases:
- Author photo day — the AI image of "person at desk writing" isn't you, you have a real photo, use the real one
- Your actual book cover for a cover-reveal post
- A screenshot of a fan tweet you want to feature
- A photo from an event you're at
For carousels, Upload-Your-Own works per-slide too — see carousel modes explained.
The video-missing badge
When a post has a video that should have been generated but the video engine failed (timeout, rate-limit, empty pollen balance), the studio shows the cover frame with a plain-language note that the video itself didn't finish — it never poses as finished art.
Two ways to fix:
- Regenerate — optionally on a different video engine via the per-post picker
- Upload your own video file
Regenerating a post that already has media genuinely regenerates it, by the way — the studio waits for the new media to land before calling it done.
What gets persisted
The per-post engine choice applies to that regeneration run only — your campaign and Settings defaults never silently change. What is recorded is the provenance: every post remembers which provider and model actually produced its image and its video, and the studio shows it under the preview.
Common mistakes
1. Reaching for the engine swap first. Try Regenerate (same provider) once before swapping. AI variance usually solves the problem with the provider you already configured.
2. Swapping providers without changing the prompt. The caption is the prompt. If "moody coastal scene" produces wrong mood on Magnific, "moody coastal scene at dusk, slate-gray water, single lighthouse" might produce right mood on Magnific without needing to swap providers.
3. Uploading manually to cover for chronically-bad AI output. If you're uploading for half the campaign, either the provider isn't right for your aesthetic, or your persona's Image Style Guide is too vague. Fix the upstream issue, not each individual post.
4. Forgetting Skip exists. When AI keeps failing one specific post for unclear reasons, skipping is faster than fighting it. The batch survives one missed post.
What to do next
- For more on the providers themselves, Magnific (formerly Freepik), explained.
- For the carousel-specific equivalents (per-panel regenerate and the panel text editor), carousel modes explained.
- For text providers (separate from media), OpenAI vs Claude vs Gemini.
- For the full campaign workflow, 14 days of content.
Read next
Running an image-only campaign
The lowest-lift batch in the tool: a run of single images in your brand look, captioned per platform, dropped into your queues. Here's the full walkthrough, Look check to publish.
Sep 5, 2026 · 5 min read
Carousel modes, explained: per-slide vs single-repeated, slide caps, and per-slide overlays
Carousels stopped being one-size-fits-all. Two image strategies, four recipes, per-platform caps, and a real panel-text editor — here's when to pick what.
May 10, 2026 · 8 min read
Magnific (formerly Freepik), explained — and the other AI media providers we support
Three AI media providers handle the visual side of your social content: Magnific, fal.ai, and Google's Gemini. Here's what each one does, when to pick which, and how to set them up.
May 10, 2026 · 7 min read