# Analyse: Workflows, Vorlagen & Prompts

> Extrahiert aus dem Telegram-Chat "Elite - AI OFM" (43.141+ Nachrichten, 783 User)
> Stand: 02.05.2026

---

## 1. US-SIM / US-Location: Pflicht oder nicht?

**Der Chat ist gespalten. Es gibt zwei Lager:**

### Pro-US-SIM (Mehrheit)
- Algorithmus schiebt Content NUR ins Herkunftsland der SIM
- Deutsche SIM = deutsches Publikum = weniger Monetarisierung
- "Tello eSIM $5/Monat" wird oft genannt

### Contra-US-SIM (Minderheit, aber erfolgreich)
> *"my irl friend. doesn't do ofm. knows nothing about ai other than chatgpt. doesn't even use IG and this man has done over 300k views his first week milking the same format over and over and did $1.6k in his first 5 days — all tier1 countries"* — phonefarmguy

> *"I have 2.5m views on instagram in the last 7 days (full usa/europe audience)"* — Lina

> *"posting high quality reels after interacting with IG users (warmup) also works. I got 284k views on my first 5-6 reels with 0 followers. literally 0"* — Kaloutsidis

**Fazit:** US-SIM = sicherer, konsistenter. Aber **Content ist entscheidender als die SIM**. Wenn dein Format viral geht, pushed der Algo auch ohne US-SIM. Deutsches/europäisches WLAN = schlecht — immer mobile Daten nutzen.

---

## 2. ComfyUI-Workflows: Woher bekommen?

### Konkrete Quellen aus dem Chat

| Quelle | Was | Link |
|--------|-----|------|
| **CivitAI** | Free NSFW Workflows | `civitai.com/models/86788/real-pussy-lily` |
| **CivitAI + ZIT** | "Realistischer als NBP" | `civitai.com/models/2230417/zit-breast-slider` |
| **Discord** | NSFW Workflow + Tutorial + FREE Nano Banana Pro | `discord.gg/qKnyHgTAGr` |
| **HuggingFace** | NSFW Video Uncensored | `huggingface.co/spaces/Heartsync/NSFW-Uncensored-video2` |
| **GitHub** | SynthID Bypass | `github.com/00quebec/Synthid-Bypass` |
| **GitHub** | Wavespeed Desktop (ComfyUI-Alternative) | `github.com/WaveSpeedAI/wavespeed-desktop/releases` |
| **GitHub** | Instagram Deeplink-Tool (Open Source) | `github.com/Sententional/open-in-external-browser` |

> *"Bro I got 15 workflows all from github YouTube and telegram"* — Wh

> *"Most people aren't using 1 'magic' ComfyUI workflow anymore, they just use few proven setups"* — Success

### xstus Face-Swap-Stack (aus msg-104 + Face Swap Tools)

```
Base Model: Flux Dev FP8 (17GB)
Face Consistency: PuLID Flux v0.9.1 + InsightFace + Eva CLIP
Realism LoRAs: XLabs Realism, Photorealistic Skin
Content: Pinterest-Bilder → FaceSwap → Pose ändern → Struktur anpassen
```

---

## 3. Kling-Videos: How-To

**Motion Control = das Hauptformat.**

> *"there is many types of motion control u can do"*

### Pricing
- **Higgsfield:** $125 Plan = 1200 Credits = 800 Sekunden Motion Control ($0.06/Sekunde)
- **Kling API:** Ähnlicher Preis, direkter

### Workflow (aus mehreren Nachrichten zusammengesetzt)

```
1. Video-Vorlage aufnehmen (selbst filmen oder Pinterest/TikTok)
2. Eigenes AI-Model-Gesicht via FaceSwap drauf
3. Kling Motion Control → Video generieren
4. Erstes Frame löschen (enthält SynthID-Wasserzeichen)
   → github.com/00quebec/Synthid-Bypass
5. Audio ersetzen (Trending Sound aus IG-Reels-Library)
6. CapCut → Zoom/Final-Cut
```

> *"quick question when uploading Kling motion control vids. u shud always delete the first frame of the video cuz it has SynthID on it?"* — appal

> *"when you create reels with motion control, is it neccesary to change/spoof the audio from original video?"* — Michal

### Beispiel-Prompt für Kling Motion Control

> *"refer to the beautiful black hair woman, she is taking a close-up selfie against a plain wall. she has natural green eyes and slight freckles across her nose. realistic, high-quality image."* — xstu

---

## 4. Prompts: Wie erstellen?

### peters Methode (Reverse-Engineering via Grok)

> *"Find an image you want to replicate. Put it into Grok and paste: You are a master prompt writer for seedream. I am going to give you an image and you are going to create a carbon copy. I don't want any extra text, only the prompt."* — peter

### xstus Prompt-Formel

```
"refer to the [Beschreibung], she is [Pose/Aktion] against [Hintergrund].
she has [Augenfarbe] and [Merkmal]. realistic, high-quality image."
```

### Wh's Base-Model-Ansatz für Face-Konsistenz
> Flux Dev FP8 + PuLID Flux v0.9.1 + XLabs Realism LoRA + Photorealistic Skin LoRA
> → Prompt simple halten, LoRAs machen die Arbeit

### Häufigste Prompt-Fehler
- Zu komplexe Prompts → AI produziert Unsinn
- *"I try to use seedream and it's so shit, generating me butterflies as if it were the pussy"* — HQ
- **Lösung:** Erst Base-Image generieren, DANN verfeinern. Nicht alles in einen Prompt packen.

---

## 5. Vorlagen-Videos / "Formate klonen"

**Das ist DER Kern-Tipp aus dem gesamten Chat.** Alle erfolgreichen Creator machen dasselbe:

> *"milking the same format over and over"* — phonefarmguy (über seinen Freund mit 300k Views in Woche 1)

> *"my first reel went on 120k and in the next 5 reels I was on like 600k — 60% of my reels were 100k or 200k+"* — David

### Die Strategie
1. **5-6 verschiedene Formate testen** (Motion Control, Carousel, Talking Head, Thirst Trap)
2. **Best-Performer nach 2 Wochen identifizieren** (Views + Share-Rate)
3. **Dieses Format KLONEN.** Nicht neu erfinden. Immer wieder dasselbe Schema.
4. **Share-Rate berechnen:** `Shares ÷ Views`. Unter 2% → Format wechseln.

### Was "Format klonen" konkret bedeutet
- Gleicher Video-Aufbau (z.B. "Model läuft auf Kamera zu → Close-up → Text-Overlay")
- Gleiche Caption-Struktur (Share-Call-to-Action + Emojis)
- Gleicher Posting-Zeitpunkt
- Gleiche Hashtag-Gruppen
- Nur das Model/Outfit/Location variieren

> *"optimize for sharability. CEO said it himself: highest weighted actions is views, likes, shares. But something with high views and not enough weight from likes and especially shares will be slow dripped. Higher share rate earlier on = higher reach velocity."* — phonefarmguy

---

## Bonus: Viral-Hebel (aktuelle Insights)

> *"so the goal is to get like 200 followers to ig from different platforms or boosts and then algo will start to push your reels"* — xstu

> *"Thats how I do too. Basic motion control kinda vibe.. but I noticed that if the person goes out of frame, the video tanks. Keep the face centered the whole time."* — Aylen

> *"I stoped all the stupid dancing shit and motion control stuff and started doing carousels with thirst-trap captions. Way better engagement, less effort."* — Aylen

> *"on higgsfield you can use soul 2.0 to create the images then use motion control to make the videos"* — Chris

---

*Quelle: Telegram Chat "Elite - AI OFM" — kuratiert aus 43.141+ Nachrichten*
*Generiert: 02.05.2026*