I Made a Viral BeamNG-Style Video in 5 Minutes Using Only AI

What the video claims / analyzes (viral BeamNG-style car-crash niche)

• The speaker says a fast-growing YouTube niche is AI-generated “car crash gameplay” videos (e.g., cars vs. water pits, bollards, and other obstacles).
• They cite extreme performance examples: videos with hundreds of millions of views, plus channels built around this style.
• Core “exposure” claim: most of these “gameplay” videos are not real gameplay—they are fully AI-generated by orchestrating prompts and image/video generation tools.

The overall premise: these channels simulate continuous car-crash “gameplay,” but the results are produced via AI workflows rather than actual in-game footage or physics runs.

---

The “system” tutorial (step-by-step AI workflow)

Prerequisite: one master prompt

• Use one master prompt (the speaker says they provide it “for free” via a link in comments) that generates everything needed for the niche.

---

Step 1 — Generate viral topics using Claude AI

1. Copy the master prompt and open Claude AI (speaker claims it performs best after testing multiple AI tools).
2. Paste the prompt without changing a word.
3. The output includes 10 ready-made video topics modeled on what already gets millions of views.
4. Topic template examples mentioned:
   • “cars versus giant water pit”
   • “cars versus chain bollards”
5. If the results don’t work, regenerate to get more topic ideas.

---

Step 2 — Convert the chosen topic into a scene-by-scene plan

1. Copy the selected topic prompt and paste it back into the same Claude prompt.
2. Claude generates:
   • the video concept / storyline
   • image prompts for each scene
   • video prompts for each scene
3. The speaker includes a simple “lock it in” action by typing “okay”.

---

Step 3 — Generate video clips from images using Flow AI

• Use Flow AI (speaker claims it’s 100% free with daily free credits: “50 credits daily”).

In Flow

1. Create a new project.
2. Paste the first image prompt from Claude.
3. Set aspect ratio to 9:16 (vertical Shorts).
4. Select “Nano Banana 2” model (speaker claims it yields cleaner cars, better physics, and more realistic crashes).
5. Generate the required images.

Animating images into clips

1. Use the “three dots” menu and “add to prompt” to bring the image into the prompt box.
2. Copy the matching video prompt Claude provides.
3. Switch Flow from image to video.
4. Set number of videos = 1.
5. Select “Omni” model.
6. Generate (speaker says it may take 10–20 seconds).

---

Continuity trick (making clips connect like continuous gameplay)

After generating a clip, the speaker claims the key for seamless continuation is:

1. Click the clip’s three dots
2. Choose “add to prompt”
3. This makes Flow use the last frame of the previous clip as the starting point for the next one.

Then:

• Copy the next video prompt from Claude
• Paste it into Flow
• Generate again

Result claimed: the clips maintain the same road/cars and feel continuous across multiple segments.

---

Problem + fix (when free credits run out)

Issue

• As you generate more clips, free credits may run out, causing “add to prompt” to fail.

Credit-agnostic workaround

1. Open the last generated video.
2. Use the Flow editor timeline to select the very last frame.
3. Choose “copy video frame”.
4. Paste that frame into the prompt box using Ctrl+V.
5. Paste the next video prompt and generate again.

Extra fallback

• If credits hit zero, open Flow in a fresh browser to get another “50 free credits”.
• Continue by pasting the last frame + the next prompt.

---

Editing + export

1. Download clips from Flow.
2. Use an editor like CapCut:
   • import clips in the Claude-planned order
   • place them sequentially on the timeline
   • add transitions/effects
3. The speaker emphasizes that the last frame trick helps clips feel like one continuous gameplay recording.

---

The “viral quality control” advice (upload gate)

The speaker claims a decisive difference between low-performing and high-performing channels is:

• Never upload a clip that looks even slightly off.

Quality checks mentioned:

• broken scenes
• weird physics
• glitchy cars

If something’s wrong:

• regenerate only that one clip (speaker describes the cost/time as about 2 minutes).

Reason given:

• viewers swipe away quickly; bad clips cause the algorithm to bury the video early.

---

Main speakers / sources (end)

• Main speaker: the YouTube video narrator (single person; not otherwise identified).
• AI tools used as sources in the workflow:
  • Claude AI: idea/topic + scene/video prompt generation
  • Flow AI: image-to-video and clip generation