powertokens/powertokens-media

Create English and Chinese (Mandarin/Cantonese) voiceovers and dubbed versions of a video with MiniMax speech through PowerTokens — translate and adapt the script for timing, pick matching voices, generate per-line audio, swap captions, and assemble EN and ZH cuts of the same visuals. Use when the user wants bilingual, Chinese, English, localized or dubbed voiceover, 中文配音, 英文配音, or the same ad in two languages.

Generate images from text, or edit and restyle images from reference pictures, with ByteDance Seedream through the PowerTokens API, and save them locally. Use when the user asks to create, generate or draw an image, picture, poster, product shot, thumbnail, illustration or storyboard frame, or names Seedream.

Pick the right PowerTokens model (default Seedance 2.0 Mini, plus Wan 3.0, Wan 3.0 Prime, Seedance 2.0 Fast, Seedance 2.5, Kling v3, Seedream, MiniMax speech) for a video, image or voice job by quality, speed, price, duration, aspect ratio, resolution and input type (text-to-video, image-to-video, reference images). Use before any PowerTokens generation, whenever the user asks "which model", compares models, has a budget, needs 1080p/4K, long clips, vertical video, or character consistency.

Turn a JSON shot list into a finished video with PowerTokens — Seedream keyframes, batched Wan/Seedance/Kling clips, MiniMax voiceover, then ffmpeg assembly with captions — with a cost estimate and user approval before spending, and resume after interruptions without paying twice. Use for any multi-shot job (ads, short dramas, explainers, storyboards, batches of variants), whenever the user wants more than one clip, a storyboard rendered, or a final edited video.

Produce a short vertical product ad (TikTok, Reels, Shorts, paid social, 9:16, 10-30 s) from a product URL and/or product photo — extract a claim sheet, get user approval, then hook/script, storyboard, Seedream keyframes that keep the real product, image-to-video clips, voiceover, captions and an assembled MP4, through PowerTokens. Third-party consumer brands are forbidden as props (no MacBook, logos, or other brand-identifiable devices) unless that brand is the SKU or the user approved it. Use when the user wants a product ad, promo video, e-commerce or Amazon/Shopify video, launch teaser, pastes a product link, or says "turn this product photo into a video ad".

Product-consistency QC for PowerTokens product ads — generic product bible (category → subtype → parts BOM), prompt hard constraints, pipeline pre-flight bible gate, post-keyframe vision rubric (MiniMax-M3 or glm-5.x via chat), optional human approve before i2v. Third-party brand props fail logo_policy_ok (MacBook, logos, other brand-identifiable devices) unless that brand is the SKU or the user approved it. Use for any physical SKU (bottle, bag, sneaker, device, cosmetics, …), product consistency, product bible, 产品质检, 产品一致性, or lock sheet — not thermos-only.

Write and fix prompts for PowerTokens video, image and voice models (Wan 3.0, Seedance 2.0/2.5, Kling v3, Seedream, MiniMax speech) — shot structure, camera language, timing, dialogue and audio cues, reference-image wording, and common failure fixes. Use whenever you write a prompt for any PowerTokens generation, when a result looks wrong (morphing, extra fingers, garbled text, wrong motion, drifting faces), or when the user asks for a cinematic, ad, UGC or short-drama look.

Write and produce AI short dramas and multi-shot story videos (vertical micro-drama episodes, cliffhangers, narrative shorts, animated stories) with consistent characters across shots and episodes — beat sheet, shot list with 1-3 s cuts, character reference sheets reused in every shot, Seedance/Wan clips with dialogue, captions and assembly via PowerTokens. Use when the user wants a short drama, micro-drama, web series episode, story video, storyboard to video, or recurring characters.

Turn one PowerTokens video or campaign into platform variants and A/B tests — 9:16, 1:1, 16:9 crops or native regenerations, alternative hooks, caption and CTA variants, length cut-downs (6 s, 15 s, 30 s) — with cost estimates before any regeneration. Use when the user wants versions for TikTok, Reels, Shorts, YouTube, Instagram feed, X/LinkedIn, ad A/B testing, or "make this square/landscape".

Turn text into natural speech (text-to-speech, voiceover, narration) with MiniMax speech models through the PowerTokens API, saving an MP3 or WAV locally. Use when the user asks for a voiceover, narration, TTS, read-aloud audio, or a spoken version of a script.

Turn a travel or photography still into a 3:4 vertical 「第二世界」photo-collage poster (faithful top photo + warm paper bottom with a photo-specific invented scene, optional tiny line figures, one short handwritten English caption) via Seedream on PowerTokens. Use when the user asks for a travel poster, second-world collage, 第二世界海报, 旅行海报, 拼贴海报, photo collage poster, indie magazine travel print, or to "make this photo into a collage poster".

Make a phone-native vertical UGC ad — testimonial, unboxing, problem/solution, POV demo, or three-reasons — with a fictional persona or hands-only b-roll, short spoken lines, burned-in captions, and hook variants, through PowerTokens. Use when the user asks for UGC, creator-style, talking-head, influencer-style, unboxing, testimonial, or a TikTok ad that looks filmed on a phone. NOT for a packshot commercial or product-URL claim-sheet ad (powertokens-product-ad), a multi-shot story episode (powertokens-short-drama), a single clip with no ad beats (powertokens-video), or a desktop GUI editor.

Generate AI videos (text-to-video and image-to-video) with Seedance, Wan and Kling through the PowerTokens API, and save the MP4 locally. Use when the user asks to make, generate or animate a video, clip, shot, b-roll, product video, or to turn an image into a video, or names Seedance, Wan or Kling.