# AI video for podcast clips — vertical cuts, loops and covers

> Visuals for an audio-first show on eroq — vertical cuts, atmosphere loops and episode covers, plus where transcription stops and your editor starts.

Published 2026-09-03 · eroq.ai — canonical: https://eroq.ai/blog/ai-video-for-podcast-clips


An audio show has a distribution problem that has nothing to do with audio. The episode is fine. The clips that are supposed to sell the episode need pictures, and you did not sign up for pictures. Most podcasters end up with the same three options — a waveform on a colored rectangle, a camera pointed at a microphone, or nothing — and generated video is a fourth one, provided you know exactly which parts of the job it does and which parts it does not touch.

## What eroq will not do to your episode

Be clear about this before you plan a workflow around it. **eroq does not take a podcast file and hand you captioned vertical clips.** There is no pipeline that ingests an MP3, finds the good moment, burns subtitles and exports a Reel. Nothing in eroq trims, cuts, captions, transitions or grades — it renders assets and stops.

What does exist is the transcription half. [Scribe One](/models/eroq-scribe-one) is speech-to-text at 5 credits per request, taking audio up to 8 MB with automatic language detection. It gives you the words. Picking the sixty seconds worth clipping, laying the subtitles over the picture and exporting the vertical file all happen in your editor, exactly as they do today.

So the honest division of labor for an audio-first show is this: eroq makes the **pictures**, Scribe One makes the **text**, and your editor marries them to the audio you already have.

## Three visual assets per episode

Per episode, three things are worth rendering:

1. **An atmosphere loop** — the bed that plays under the pull quote while the guest talks. A rain-streaked window, a desk lamp at 2am, a highway at dusk. It has nothing to do with the topic and everything to do with not being a waveform.
2. **Three or four vertical cuts' worth of visuals** — one short clip per moment you plan to clip, in 9:16, so the same episode yields several posts that do not look like each other.
3. **An episode cover** — one square image that is recognizably part of your show's family.

The loop is the highest-leverage asset because it is reusable. A show with six loops in rotation never has to render anything for a standard episode again.

## The atmosphere loop prompt

The [video studio](/studio/video) takes one flowing paragraph — subject, motion, camera, light, mood — and folds the director's rack in for you:

> Rain running down a dark office window at night, static locked-off frame as headlights smear into bokeh on the street below and a desk lamp glows out of focus in the foreground, anamorphic lens at f/1.4, neon noir palette, practicals only, calm tempo, arthouse look, close-up.

Static is the right camera move here and it is worth saying why. A loop plays under speech for thirty seconds; a push-in that ends somewhere is a shot, but a frame that barely moves is a bed. Handheld and Orbit are for the cut-aways, not for the thing your voice sits on.

Render the loop at 10 seconds and let your editor repeat it. Ten seconds of rain on glass cuts back to its own first frame without anyone noticing.

## Episode covers with Image One

[Image One](/models/eroq-image-one) is the photoreal flagship at 1024×1024 for 10 credits an image, with batches of 1 to 4, reference photos, a negative prompt and CFG control. Batch four variants of a cover, keep one, and you have spent 40 credits on a decision that used to be a Canva evening.

Two practical notes. Keep "text, watermark, logo" in the negative prompt — generated lettering is always almost-words, and you want your title typed in your own design tool anyway. And there is no upscaler in eroq, so 1024×1024 is what you get from the model; if your host wants a larger cover, place the render in your image editor and finish it there.

If the show is illustrated rather than photographic, Image Anime handles stylized work and Image Art follows long descriptive prompts in a painterly register, both at the same 10 credits. Set the aspect ratio in [the image studio](/studio/image) — 1:1 for the cover, 9:16 for the story frame, 16:9 for the YouTube thumbnail plate.

## What one episode costs

One episode, four vertical clips at 5 seconds on [Seedance 1.0 Lite](/models/seedance-1-lite) at 60 credits, one 5-second atmosphere loop on [Motion One](/models/eroq-motion-one) at 100, four cover variants on Image One at 10, and an hour of audio transcribed in however many chunks fit under the 8 MB per-request limit at your export bitrate — call it four — at 5 credits each:

- Vertical clips — 4 × 60 = **240**
- Atmosphere loop — 1 × 100 = **100**
- Cover variants — 4 × 10 = **40**
- Transcription — 4 × 5 = **20**

Total **400 credits** an episode. A weekly show is 4 × 400 = **1,600 credits a month**, which fits inside [Hobby](/pricing) at $15 for 1,800 with 200 to spare.

Here is where the arithmetic bites: render the loop at 10 seconds instead of 5 and Motion One charges 180 rather than 100, so the episode is 480 and the month is 1,920 — 120 credits over the Hobby allowance. The fix is not a bigger plan, it is rendering the loop once and reusing it for a quarter. A loop is not per-episode content.

If the show is fortnightly, the same 400-credit episode costs 800 a month, and the 50 free credits a new account starts with are enough to render five cover variants and decide whether you like the look before paying anything. They are not enough for a clip — the cheapest 5-second render is 60 credits on Seedance 1.0 Lite and 100 on Motion One, the engine a free account has — so spend them on images.

## The cadence that makes it worth it

Batch. Render six atmosphere loops in one sitting, four covers per month in another, and stop touching the video studio between episodes. Every render is saved in your library with its recipe, and Remix reloads that recipe into the tool, so "the rain one, but snow" is a two-word edit rather than a rewrite. The audio-side arithmetic — narration, character billing, the two voice engines — is covered on the [audiobooks and podcasts use case](/use-cases/audiobooks-podcasts).

## FAQ

### Can eroq turn my episode into captioned clips automatically?

No. Scribe One transcribes your audio at 5 credits per request for files up to 8 MB, and that is where eroq's involvement in your timeline ends. Choosing the clip, burning captions and exporting vertical happen in your editor.

### What should I use for the visuals under a talking clip?

A near-static atmosphere loop, rendered once and reused. Motion One is 100 credits for 5 seconds and 180 for 10 on any account, Seedance 1.0 Lite is 60 for 5 seconds on Hobby and up. Movement competes with speech, so keep the camera still.

### Can I generate a picture of my guest?

No. Real identifiable people without consent are banned by policy, and generated people are fictional strangers. Use the guest's own photo, or keep the frame on atmosphere.

Recording this week? [Open the video studio](/studio/video), render two atmosphere loops, and never post a waveform again.