# Miss Lily — Claude's zero-budget video production brief Prepared 8 October 2026. This is an instruction brief, not a claim that any videos have been rendered. The attached lesson bank contains the approved narration and scene directions for all 50 mini lessons. You already have access to the Miss Lily character assets and character traits; locate and use those assets. ## What to do Act as the producer and operator. Use your browser/computer tools and local file tools to test eligible free video generators, choose the best usable route, and make actual downloadable lesson videos. Do not just return prompts or tell the user how to do the work. Begin with **Lesson 01, Priya: Client की call पर report भेजने का time कैसे पूछें?**, the latest approved concise pilot. Older packs called a different lesson “Video 4”; that legacy label does not override this pilot. The audience is Hindi-speaking Indian adults who know some English words but freeze when speaking. They need one small useful line for a recognisable situation, delivered without shame. The commercial goal is useful mini lessons inside the Miss Lily trial and monthly subscription. Preserve that intent; do not turn these into generic motivational shorts. Use **$0 / ₹0 monetary spend**. Existing free credits may be consumed for this task. Reuse my existing signed-in accounts and normal password-manager sign-in flows on official services. Do not extract browser cookies, put credentials in prompts, log credentials, or send them to unrelated sites. No subscriptions, credit purchases, card-backed trials, paid APIs, upgrades, or recurring commitments. Skip an unavailable or paid feature and continue with another eligible option. Do not create extra accounts to obtain credits or bypass geographic/account restrictions. If your browser product cannot create accounts, skip unsigned-in services. Bundle genuine login/OTP/CAPTCHA or mandatory terms steps for me only when needed; continue other useful work first. Do not bypass your tool's approval requirements. Use a Claude session that has browser/computer access AND local file access. A chat-only session cannot perform this workflow. Use normal site permissions rather than disabling protection. Website access, downloads, authentication or CAPTCHAs may still require user action; do not promise unattended access to every provider. ## The production rules - Use exactly the repository presenter named for each lesson. One presenter, one consistent voice, one visual identity throughout. No stock avatar substitutions, invented characters, additional actors, background people or second speakers. References to a client, interviewer or teacher are narrated hypotheticals, never a cast member. Use the repository look and traits; inspect the actual reference before generating. Do not assume that an existing image asset grants a new voice-cloning permission or supplies an audio sample. - Start on the presenter speaking the concise situation immediately. No greeting, logo intro, silent establishing shot, or music opening. Preserve the lesson's hook, freeze and concrete English solution. - Speak the supplied narration exactly, in order. Hindi stays in Devanagari; English phrases stay in Roman script. Scene headings, direction notes and graphic instructions are NOT spoken. Do not translate, summarise, add an exchange, replace the lesson or remove the repeated connected example to satisfy a tool limit. - Use natural Hindi/Hinglish, slightly slower clear English examples, warm steady eye contact and restrained gestures. No overacted panic, mockery, exaggerated celebration, fluency guarantees or guaranteed interview outcomes. The selected repository voice must be compatible with these traits; verify the actual audio rather than trusting a language badge. - Keep the presenter visible for every English teaching example, roughly 80% or more of the whole video. Use short object-only B-roll only under the existing Hindi explanation; same voice continues. No faces, people, hands, body parts, silhouettes, reflections of people, portraits or mannequins in B-roll. No separate B-roll voice. - Use a few simple editable phrase cards and accurate subtitles, with no more than two readable lines at a time. Add text in the editor; do not ask a video model to render Hindi text. Keep cards away from the face. Conditional examples stay conditional; no invented reply to turn a suggestion into a guaranteed result. - No viewer repetition request, quiz, recording task, practice silence, sales CTA or extra ending. Finish after the last supplied sentence. Use a plain clean background and no music unless an existing approved, properly licensed track is available; silence beneath speech is acceptable. - Default to 9:16, 1080×1920 if freely available; accept a clearly labelled lower-resolution prototype. Never stretch landscape output. Length is set by natural spoken delivery, usually about 70–90 seconds; supplied estimates are planning estimates, not measured renders. ## Audit the actual free accounts before spending credits Check official services, prioritising whichever existing account genuinely supports a reference-led talking presenter and export. Candidates are **HeyGen, Kling AI, Dreamina/Seedance and Runway**. Seedance is a model available through platforms such as Dreamina; use the official platform, not an unverified site using the model name. Add at most one other well-known service only if the four candidates fail and its free/custom-avatar capability and usage rights are verifiable. For each account, record: official URL; signed-in availability; free credits before generating; model/features actually included; reference image support; Hindi/Hinglish audio/lip-sync capability; maximum clip duration; free downloadable export versus share link only; resolution; watermark; and permission for use in a paid subscription product. Check applicable official plan and usage terms. If rights are unclear, label the result **private evaluation only**, not publish-ready. Do not treat marketing claims or a free signup button as proof of free rendering/export. Never click a charge/upgrade button to discover whether it costs money. Research notes checked on 8 October 2026 (account UI and current terms take precedence): | Candidate | What the official source indicates | How to use it here | |---|---|---| | HeyGen | Free allowance is limited and may vary by region; official help describes up to one-minute free videos and limited Video Agent/custom-avatar access. A sharing link is not proof of free MP4 download. | First check the existing account's reference avatar, audio and export entitlements. A 70–90 second lesson needs compliant segments and assembly if a 60-second cap applies. | | Runway | The documented free plan has 125 one-time credits, limited available models and watermarked exports; free credits can be absent on some accounts. | Test reference identity/animation or object-only B-roll if included. Do not assume a silent image-to-video model can narrate a complete lesson. | | Dreamina / Seedance | Official Dreamina pages advertise free credits and Seedance models; access and available models must be checked on the actual account. | Check current credits, region, reference inputs, native Hindi audio and duration/export before generating. | | Kling AI | A current free quota was not verified from accessible official documentation. | Candidate to inspect, not a guaranteed free option. Verify account credits, reference consistency, audio/lip-sync and export. | Official sources: - HeyGen: https://help.heygen.com/en/articles/15125761-heygen-credit-based-pricing-plans-subscriptions-explained and https://www.heygen.com/en-gb/pricing - Runway: https://help.runwayml.com/hc/en-us/articles/50404627334547-Free-plan-details and https://help.runwayml.com/hc/en-us/articles/40521953538451-Where-are-my-Free-Credits - Dreamina: https://dreamina.capcut.com/tools/seedance-2-0 and https://dreamina.capcut.com/resource/seedance-1-5-pro - Kling: https://klingai.com/ - Claude browser permissions: https://support.claude.com/en/articles/12902446-claude-in-chrome-permissions-guide - ElevenLabs licensing: https://help.elevenlabs.io/hc/en-us/articles/13313564601361-Can-I-publish-the-content-I-generate-on-the-platform **Audio matters:** reuse approved existing voice recordings/audio assets if suitable. Otherwise choose an eligible native voice workflow whose identity and usage rights are appropriate. Do not recommend ElevenLabs free-plan audio for published Miss Lily subscription lessons: its official help excludes commercial use and a later upgrade does not retroactively license free-generated audio. Do not substitute an unrelated voice just to obtain a moving mouth. If no permitted voice route exists, save the visual tests and report the exact audio blocker. ## Run a small fair comparison, then complete the pilot 1. Inventory the existing assets and match Priya's approved reference and voice/traits. Preserve the originals. Record asset identifiers/paths locally; never publish login details or private account screens. 2. Audit the candidate accounts above. Test **up to three genuinely eligible providers**, one attempt per provider, plus at most one retry to fix a specific technical defect. Do not spend credits repeatedly tuning style. When fewer than three qualify, compare the eligible ones and record why the others were skipped. 3. Use the same Priya reference, framing and spoken excerpt for every talking-presenter probe: **“I couldn’t hear you clearly.”** followed by **“यानी—आपकी बात साफ़ सुनाई नहीं दी।”** Aim for roughly 8–10 seconds at natural pace, adjusting clip duration rather than speeding up speech. This is a contiguous excerpt of Lesson 01's teaching section, a test clip only. Add no other spoken words. Prefer reusing the same approved audio file across tools that support audio-driven lip-sync. If native voices differ, label the comparison as different voice routes rather than pretending audio was controlled. 4. Compare identity, mouth/audio timing, Hindi pronunciation, English pronunciation, stability, export quality, watermark, cost and usage rights. Watch and listen to every downloaded result. A model name or a finished job status is not a passing result. Record observed defects. Pick the best eligible route yourself; do not ask the user to choose routine settings. 5. Make the **entire Lesson 01**, with all six scenes and exact narration. Generate the continuous speech track first where possible, then derive segment timing from it. Split at natural sentence boundaries to respect each tool's clip limit; leave small trim handles, preserve the same reference/voice/look and reassemble with local free tools such as an already available FFmpeg. A scene may need several generator jobs. Do not paste the whole 90-second production brief into a 5-second image-to-video field. 6. Add the short object-only cutaways and editable phrase cards from the scene plan. Use simple local graphics when free video credits are better spent on the presenter. Generate only B-roll that adds value; static object shots or restrained pans are acceptable. Keep speech continuous; never burn text into a generated scene. 7. Export and perform full QA. If it passes, continue with other lessons in numerical order using the same workflow while genuinely free quota remains. Check each lesson's presenter: do not reuse Priya for a Miss Lily or Rhea lesson. Do not request approval again for the already approved scripts. Do not claim all 50 can be rendered within finite trial quotas. Stop paid/blocked routes, preserve checkpoints, and report exactly which lessons are done and what free capacity remains. Do not silently schedule future runs. If no eligible service can produce a complete talking-presenter pilot at zero cost, deliver the actual provider probes and, only if approved voice audio is available, a clearly labelled **still-presenter narrated prototype** with the full script, cutaways and cards. It must use the approved portrait and retain every word. Do not present this fallback as a lip-synced avatar result. If neither route is possible, return concrete completed work and the specific blocker, not a fabricated video or an instruction-only success claim. ## How to convert each scene into tool inputs Use the lesson bank below as the source of truth. The directions are provider-neutral even where inherited editorial cues mention a presenter or A-roll. For an agent-based editor, provide the global rules plus just the selected lesson. For a short video model, send only the relevant visual shot prompt/reference and its compatible audio input; put narration in the actual speech/audio field, not solely in a visual-description prompt. **Talking-presenter visual prompt template** > Vertical 9:16 educational presenter shot using the attached approved [PRESENTER] reference asset. Preserve the exact face, age, hairstyle, outfit and repository visual identity. One presenter only, chest-up, eye-level fixed camera, clean softly lit [repository-compatible setting]. Warm natural delivery and small restrained gestures matching [repository traits]. Accurate mouth timing to the supplied speech/audio where the selected model supports it. No other people, voices, scene dialogue, background faces, props obscuring the face, camera shake or identity drift. No generated writing, captions, logos or music. Maintain the same framing and lighting across segments. The attached narration is spoken exactly; production instructions are not spoken. Fill the placeholders from the ACTUAL repository, rather than inventing appearance details. For the pilot, use Priya's concise, reassuring professional delivery and a simple clean office-compatible background. When the model cannot accept speech/audio, omit the lip-sync claim and label it an animation-only probe. **Object-only cutaway prompt template** > Vertical 9:16, approximately two seconds usable duration. [Exact object-only setting from the selected scene's B-roll cue]. Quiet natural lighting, locked camera or a very slow subtle push-in. No people, hands, body parts, faces, silhouettes, portraits, mannequins or human reflections. No readable text, logos or generated captions. No dialogue, speech or music. This shot will be placed beneath the existing presenter narration; the narrator is not replaced. **Cards/subtitles:** create the exact phrases specified in the selected scene as editable overlays. Read timing from the final audio; do not guess subtitle timecodes before the audio exists. Keep English examples on presenter shots. Verify rendered Devanagari glyphs and export fonts. Do not use bold Markdown markers or quotation delimiters as spoken words. ## Files to leave behind and final report Create a local `Miss-Lily-Free-Video-Tests/` folder. Include: - `00-RESULTS.md`: tested providers, skipped reasons, measured QA, winner, actual completed lessons, zero monetary spend confirmation, free credits used/remaining, commercial-use status with dated source links. - `01-ACCOUNT-CAPABILITIES.csv`: the account audit above, with no credentials, email screenshots or personal details. - `02-CONTINUATION.md`: completed/queued lesson IDs, job IDs, remaining quotas and precise resume step. Avoid duplicate generations on resume. - `probes/`: actual downloadable test videos, labelled by provider and voice route. Do not call these full lessons. - `lessons/01/`: full exported MP4, SRT/VTT subtitles, narration text, audio file if available, exact submitted prompts and segment/assembly manifest. Repeat per completed lesson. Preserve watermarks required by the free plan; never crop/remove them to disguise provenance. QA requires: approved identity and correct presenter; all narration present exactly once in its scripted sequence (including intentional repeats); intelligible Hindi/English; appropriate pronunciation and lip timing; no second people or voices; correct English cards; no missing audio/gaps/overlaps or abruptly cut sentences; readable mobile captions; downloadable playable MP4 with measured duration/resolution; honest watermark and rights status. Watch/listen to the final assembled file, not just component previews. If a subjective point cannot be assessed with your tools, disclose it and attach the preview rather than inventing a score. Return links to the actual files, a compact comparison, the recommended workflow and the remaining blocker if any. Continue routine execution without asking for choices already answered in this brief. Do not email anyone, publish the generated videos externally or modify production Miss Lily lessons unless separately instructed. --- # Begin with this approved pilot ## Lesson 01 — Client की call पर report भेजने का time कैसे पूछें? Presenter: **Priya** · Track: Workplace · Estimated 70–85 seconds (unrendered) Learning goal: Clarify a missed deadline on a client call The six **Spoken narration** blocks below, concatenated in order with a blank line between them, are the complete narration. All other text is direction only. Do not read direction headings aloud. ### Scene 1 — Set up the client's call and report request **Spoken narration — exact:** सोचिए, आपके client ने आपको फोन किया। उन्हें आज एक report चाहिए, लेकिन आवाज़ कटने से time सुनाई नहीं दिया। **Visual, delivery and editable graphic cues — not spoken:** #### A-roll direction Open on Priya looking into the lens. She guides the viewer into a simple imagined situation, explaining who called, what they need, and what detail was missed. Stay on her for the opening sentence. She narrates all of it without a caller voice, phone role-play, or comic panic. #### B-roll cue On **“उन्हें आज एक report चाहिए”**, show an approximately two-second cutaway to an unoccupied office desk with a closed laptop, plain folder, and a phone lying face down. Warm daylight; no hands, reflections, faces, logos, or readable paperwork. Return to Priya before **“लेकिन आवाज़ कटने से”**. Keep her voice continuous. #### On-screen text During the report request, show the editable label **आज report भेजनी है**. On **“time सुनाई नहीं दिया”**, replace it with **कितने बजे तक?**. Keep each card simple and at most two lines. ### Scene 2 — The missed detail and the worry **Spoken narration — exact:** आप पूछना चाहते हैं—कितने बजे तक? पर English में अटककर बस कह देते हैं— “Yes, sure.” **Visual, delivery and editable graphic cues — not spoken:** #### A-roll direction Use a closer framing. Priya recognises the difficulty starting the question. Keep her visible for “Yes, sure”, with a restrained pause before it. Do not portray a humiliating client reaction or exaggerate distress. The call is still in progress; do not suggest the learner has hung up. #### Graphic cue On **“आप पूछना चाहते हैं—कितने बजे तक?”**, briefly show two small editable cards below Priya's face: **Report** and **Time?**. The first is clear; the second stays muted. Keep Priya visible and clear the cards before “Yes, sure”. Use a plain fade, no noisy glitch or audio distortion. Do not extend this short scene for a separate cutaway. #### On-screen text Show **Yes, sure** when spoken and clear it before the graphic. No extra slogans, judgement labels, or red failure marks. ### Scene 3 — Explain what was not heard **Spoken narration — exact:** ऐसे में दो छोटी lines मदद कर सकती हैं। पहले बताइए कि आवाज़ साफ़ नहीं आई— “I couldn’t hear you clearly.” यानी—आपकी बात साफ़ सुनाई नहीं दी। **Visual, delivery and editable graphic cues — not spoken:** #### A-roll direction Keep Priya visible throughout. Deliver the English phrase naturally, with clear articulation of “couldn’t”. She is demonstrating a useful line, not making an actual call. No practice gap for the viewer. #### Graphic cue Show **I couldn’t hear you clearly.** as one editable phrase card when spoken. Let it remain briefly during the Hindi meaning, then remove it. ### Scene 4 — Ask the missing question and connect the lines **Spoken narration — exact:** फिर जो जानना है, वही पूछिए— “What time do you need the report?” मतलब—report कितने बजे तक चाहिए? दोनों lines साथ सुनिए— “I couldn’t hear you clearly. What time do you need the report?” **Visual, delivery and editable graphic cues — not spoken:** #### A-roll direction Keep Priya on screen throughout. Give the full question one natural sentence break, without switching voices or acting a client response. The example shows how to clarify the missed detail during the call. #### Caption cues First show **What time do you need the report?**, wrapping to two lines if necessary. During the combined answer, replace the overlay in spoken order: 1. **I couldn’t hear you clearly.** 2. **What time do you need the report?** No full-screen cutaway while the English is being taught. ### Scene 5 — Confirm the time once it is heard **Spoken narration — exact:** Time चार बजे बताया गया हो, तो confirm कर सकते हैं— “By four today, right?” यानी—आज चार बजे तक, सही? इस तरह report भेजने का time भी clear हो सकता है। **Visual, delivery and editable graphic cues — not spoken:** #### A-roll direction Keep Priya visible for the English phrase. “Four” is a conditional sample deadline, not a real instruction to the viewer or a line supplied by another character. #### Graphic cues Show **By four today, right?** when spoken. During **“आज चार बजे तक”**, a small editable **आज · 4 बजे तक** label may replace it. Do not keep both captions at once. #### B-roll / graphic cue On **“time भी clear हो सकता है”**, show a two-second return to the simple Scene 2 graphic. Replace **Time?** with **आज · 4 बजे तक**, keeping **Report** on the other card. This represents the sample being clarified, not a guaranteed outcome. Priya's voice continues. Return to her for the close. ### Scene 6 — A small question can clear a big confusion **Spoken narration — exact:** दुबारा पूछने के लिए perfect English का इंतज़ार मत कीजिए। एक छोटा सवाल भी आपके काम की बड़ी उलझन सुलझा सकता है। **Visual, delivery and editable graphic cues — not spoken:** #### A-roll direction Finish in close-up with steady eye contact and a small, natural nod. Keep the delivery practical and reassuring. Do not overpromise fluency, approval, or professional success. #### B-roll cue No cutaway. End naturally on Priya after her final sentence. Narration integrity SHA-256: `a2903d25f1e395a9ff70dc858812b9036160c8468b2e9df6de93511e0f518f93` (UTF-8, scenes joined with two LF newlines; no trailing newline).