PROMPTS USED IN THE VIDEO
Prompt 0. Connecting Fal and Blotato
I want to connect two tools to Claude Code so they work in every project on this computer. Both are hosted servers that sign in through my browser, no API keys.
1. fal.ai for image and video generation. Server address: https://mcp.fal.ai/mcp-relay
2. Blotato for posting to Instagram and DM automations. Server address: https://mcp.blotato.com/mcp
Add both at user scope, then tell me exactly what to do next to sign in. Keep it to the next step only.
Prompt 1. Ground Rules
We are building an AI creator for a small matcha brand called KO Matcha, end to end: character photo, a second photo in a new place, a talking clip, post it to Instagram with Blotato, then a comment-to-DM automation. Rules for this whole session:
1. Use the fal MCP for every image and video job. Before each job, tell me the model id and the fields you are sending, in one line.
2. If any job is refused or errors, show me the exact error message verbatim and stop. Do not retry, do not switch models, do not change the image. Wait for me.
3. After each job, print the elapsed time and the cost from fal if it is returned, then save the output into ./out with a short name and tell me the file name.
4. Never print or echo any API key or token.
5. Keep your replies short. One or two lines between jobs.
Confirm you have the fal tools and the Blotato tools available, and list my connected Blotato accounts.
Prompt 2. Glossy standard prompt
Generate a 9:16 image with fal-ai/nano-banana-pro at 1K. Prompt: "Beautiful young woman influencer, flawless glowing skin, perfect makeup, long dark hair, holding an iced matcha latte, bright modern kitchen, soft studio lighting, ultra realistic, 8K, professional photography." Save it as out/glossy.png and open it.
Prompt 3. Detailed Realism Prompting
Same model, same size. Prompt: "Unedited photo taken on an iPhone front camera, vertical. A woman in her mid twenties, Filipino-American, long straight dark hair, dewy skin, a small nose stud, gym hoodie. Light everyday makeup: tinted moisturiser, brow gel, lip balm, nothing heavy. Pores still visible through it, a little shine on the nose and forehead, faint shadows under the eyes, a few flyaway hairs, slightly uneven skin tone on the cheeks. She is mid-sentence, looking at the lens, holding up an iced matcha latte in a clear plastic cup with condensation. She is in the driver seat of a parked car, seat belt off, hard sunlight through the windscreen hitting one side of her face, the rest in shade. Mild sensor grain, slight wide-angle distortion at the edges, the hand with the cup slightly soft from autofocus. Candid, not posed, not retouched, no beauty filter, no studio lighting." Save it as out/maya.png and open it next to out/glossy.png. Create also another file that is a character anchor of her showing sideview, and her in different outfits.
Give me the fal URL of out/maya.png. We will use that same image as the reference for everything that follows. Call her Maya.
Prompt 4. Another shot
Use fal-ai/nano-banana-pro/edit with Maya's image as the only reference. Prompt: "Use the woman in the reference image: same face, same nose stud, same hair, same skin. New photo: she is sitting at a window table in a small cafe in the late afternoon, holding up a small matte green tin of matcha powder with a plain cream label that reads KO MATCHA, turned toward the camera. Taken on an iPhone by a friend across the table, slightly off-centre framing, window light from behind-left with a little flare, her face half in shade, real skin texture with pores and a bit of shine, flyaway hair, mild grain, the tin slightly soft because focus is on her face. Unedited, no beauty filter, no studio lighting, candid, mid-sentence." 9:16, 1K. Save as out/maya-cafe.png and open it beside out/maya.png.
Prompt 5. Fal Seedance2.5
Now make her talk. Use bytedance/seedance-2.5/reference-to-video with Maya's image as image 1. Prompt: "[Image1] is the woman and the car. Handheld phone selfie video. She is holding the phone herself, so the frame drifts a little with her arm. She looks at the lens and says: \"Okay, day three of this matcha thing. I swear it is the only reason I am upright before nine. Comment the word SLOW and I will send you the one I use.\" Natural lip movement, she glances down at the cup once and lifts it slightly, a blink, a small head tilt. Quiet parked-car interior, faint traffic outside. The light stays constant. Real skin texture, no beauty filter, no cinematic grade, slightly warm phone colours, normal speed." 9:16, 720p, 8 seconds, audio on. Remember rule 2: if it is refused, show me the full error and stop.
Plan B if the live face happens to pass (once in nine tries): "Upload out/plan-b.png to fal and send that one to the same Seedance endpoint with the same prompt." Copy a05 or a06 into the demo folder as plan-b.png before filming.
In three plain sentences, for someone who does not code: what is this error, why does ByteDance do it, and what is the legitimate way around it? Do not suggest editing or masking the image.
Prompt 6. Switch to Kling
Send the exact same Maya image, unchanged, to fal-ai/kling-video/v3/pro/image-to-video as the start image. Prompt: the same selfie prompt as before without the "[Image1]" sentence. Duration 8, generate_audio true, negative_prompt "plastic skin, beauty filter, slow motion, cinematic, studio lighting, blur, distort, low quality". Save as out/maya-clip.mp4 and open it when it lands. Tell me the elapsed time.
Run out/maya-clip.mp4 through fal-ai/whisper and show me the transcript next to the line I asked for.
Prompt 7. Post with Blotato
Post out/maya-clip.mp4 to my Instagram account via Blotato as a Reel, publish now, not scheduled. Caption: "day three of the matcha thing and honestly I get it now. comment SLOW and I'll send you the one I use. (AI-generated creator, made for KO Matcha)”. Create a Blotato comment-to-DM automation on that live post. Trigger: any comment containing the word SLOW. Action 1: a public reply under the comment that tags the commenter and says "sent you a DM". Action 2: a DM that says "here's the one I use" with a button to komatcha link, and before the link, ask for their email and save it to the contact. Show me the automation settings before you turn it on, then turn it on.
Prompt 8 Make >15 sec clip
Make a continuous 25-second talking clip of Maya getting out of the car with her matcha. Kling 3.0 Pro only does 15 seconds per job, so do it as two clips chained on the last frame, then join them. Steps:
1. Clip A, 15 seconds. fal-ai/kling-video/v3/pro/image-to-video, start image = Maya's still (out/maya.png URL), duration "15", generate_audio true, negative_prompt "plastic skin, beauty filter, slow motion, cinematic, studio lighting, blur, distort, low quality". Prompt:
"Handheld phone selfie video, she holds the phone herself so the frame drifts with her arm. Young woman, light American accent, relaxed mid-pitch voice, unhurried. She looks at the lens and says: \"Okay, day three of this matcha thing, and I swear it is the only reason I am upright before nine. I'm gonna be honest, I did not think I'd like it. Come with me, I'll show you what I actually do with it.\" While talking she glances at the cup once, then reaches across with her free hand, pulls the door handle and pushes the car door open. The clip ends with the door open and her still in the seat, turned toward the open door, phone still on her face, mid-smile. Hard sunlight through the windscreen stays constant. Real skin texture, no beauty filter, no cinematic grade, slightly warm phone colours, normal speed."
Save as out/maya-car-a.mp4.
2. Extract the final frame of out/maya-car-a.mp4 with ffmpeg at full resolution (use -sseof -0.05 so it is the true last frame), save it as out/maya-car-a-last.png, upload it to fal and give me the URL. Open the frame so I can check her position and the door before we continue. Wait for my OK.
3. Clip B, 10 seconds. Same model, start image = that last-frame URL, duration "10", same audio and negative settings. Prompt:
"Continuation of the same handheld phone selfie, same woman, same car, same sunlight, no cut. Same young woman, light American accent, relaxed mid-pitch voice. She swings her legs out, stands up out of the car with the matcha cup in her other hand, nudges the door shut with her hip and starts walking, phone still held up on her face, and says: \"So this is the one. Comment the word SLOW and I'll send you the link, and the code I use is in the DM.\" She lifts the cup into frame on the last line. Outdoor daylight, parking lot behind her, the frame bounces a little with her steps. Real skin texture, no beauty filter, no cinematic grade, slightly warm phone colours, normal speed."
Save as out/maya-car-b.mp4.
4. Join them with ffmpeg into out/maya-car-25s.mp4: concatenate video and audio, re-encode so both parts match (same resolution, 24 fps, AAC), and apply a 0.15 second audio crossfade at the seam so the room tone does not click. Open the result. Then tell me the total elapsed time and cost of the two jobs.
