higgsfield-assist
Skill788 starsChanged 51 days agoMITView on GitHub
Use when the user asks about Higgsfield Assist (the built-in GPT-5 copilot), how to use the platform's native AI assistant, credit optimization strategies, plan selection, how to get more from fewer credits, or platform efficiency tips.
In this repository36 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- higgsfieldSkill
- higgsfield-3dSkill
- higgsfield-actingSkill
- higgsfield-appsSkill
- higgsfield-audioSkill
- higgsfield-cameraSkill
- higgsfield-canvasSkill
- higgsfield-character-designSkill
- higgsfield-cinemaSkill
- higgsfield-content-factorySkill
- higgsfield-facsSkill
- higgsfield-gpt-image-2Skill
- higgsfield-image-shotsSkill
- higgsfield-marketing-studioSkill
- higgsfield-mixed-mediaSkill
- higgsfield-modelsSkill
- higgsfield-moodboardSkill
- higgsfield-motion-designSkill
- higgsfield-motionSkill
- higgsfield-pipelineSkill
- higgsfield-promptSkill
- higgsfield-recallSkill
- higgsfield-recipesSkill
- higgsfield-repurposeSkill
- higgsfield-scene-engineSkill
- higgsfield-seedance-2-5Skill
- higgsfield-seedanceSkill
- higgsfield-seedance-vfxSkill
- higgsfield-shotlist-directorSkill
- higgsfield-soulSkill
- higgsfield-stackSkill
- higgsfield-styleSkill
- higgsfield-troubleshootSkill
- higgsfield-vibe-motionSkill
- higgsfield-workspacesSkill
The file
We split this skill into the rules it gives an agent
12 rules. About 0 are common, 11 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈11
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 80
Best workflow: Use this Claude skill to plan and structure → use Higgsfield Assist for final in-platform prompt refinement and quick …
- Very specific: line 144
Every logged generation sharpens these numbers — the logging workflow is one command (../higgsfield-recall/SKILL.md § Log the Generation …
- Distinctive: line 159
If you change the prompt, the model, AND the camera in one go, you can't learn what fixed what.
- Distinctive: line 172
Fix: Check the Apps library first.
- Distinctive: line 200
Avoid: Kling 3.0 pro/4K modes, Seedance 2.0 at 1080p/4K, Veo 3 / 3.1
- Distinctive: line 207
Use Cinema Studio for your two or three most important scenes
- Distinctive: line 229
Before burning credits on a new style or effect you haven't tried, find a community example that uses it.
- Distinctive: line 237
When a generation works well, save the complete prompt immediately.
- Distinctive: line 247
If you're using the same Soul ID character in 5 different scenes, generate them in the same session.
- Distinctive: line 258
Create a Moodboard for your project
- Distinctive: line 272
Add Moodboard + Soul ID for consistent series content
- Distinctive: line 273
Use Higgsfield Assist for rapid iteration
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Higgsfield Assist + Credit Optimization
<!-- CONTENTS — generated by scripts/build_index.py; do not hand-edit -->Contents
- Higgsfield Assist (GPT-5 Powered Copilot)
- What Assist Can Do
- How to Use Assist
- When to Use Assist vs Claude with This Skill
- Coming Features in Assist
- Credit Optimization Guide
- Understanding Credits
- Credit Cost Tiers (Approximate)
- Quote From the Ledger, Not From Vibes
- The 5 Most Common Credit Waste Patterns
- The Hero Frame Efficiency Method
- Model Selection by Budget Scenario
- Platform Efficiency Tips
- The Platform Learning Path (Credit-Efficient)
- Related skills
<!-- /CONTENTS -->Higgsfield Assist (GPT-5 Powered Copilot)
Location: higgsfield.ai/chat
Higgsfield Assist is a GPT-5 powered creative copilot built directly into the platform. It's separate from Claude — it lives inside Higgsfield's interface and is trained specifically on Higgsfield's tools, workflows, and generation patterns.
What Assist Can Do
- Generate image prompts optimized for the Soul model
- Generate prompts for viral videos in specific styles
- Navigate the platform — recommend which tool to use for a goal
- Recommend the right effects, apps, or presets for your use case
- Answer questions about features and capabilities
- Give feedback on scripts, prompts, or creative concepts
- Suggest fresh ideas when you're blocked
- Help with storyboard planning
How to Use Assist
- Click "Assistant" in the top Higgsfield header
- Select the GPT-5 model
- Ask anything — examples:
- "Generate a Soul image prompt in the style of a Helmut Newton editorial"
- "What's the best workflow to create a 30-second branded video with consistent characters?"
- "Which camera preset works best for a car chase sequence?"
- "Help me write a prompt for a product video for a skincare brand, sophisticated tone"
- Copy the generated prompt → paste into the relevant feature
When to Use Assist vs Claude with This Skill
| Use Assist for | Use this Claude skill for |
| Quick prompt generation within the platform | Building complex multi-shot workflows |
| Platform navigation questions | Structuring long-form projects |
| Viral/trend suggestions (platform-current) | Systematic MCSLA prompt construction |
| Real-time platform feature questions | Genre recipe templates and troubleshooting |
| Rapid iteration inside the Higgsfield UI | Understanding the underlying principles |
Best workflow: Use this Claude skill to plan and structure → use Higgsfield Assist for final in-platform prompt refinement and quick generation.
Coming Features in Assist
- Generate content (Image, Video, Canvas) directly inside chat
- Upload and analyze media files
- Large file analysis
- Storyboard builder from ideas
Credit Optimization Guide
Understanding Credits
| Plan | Monthly credits | Cost | Best for |
| Free | 25 | $0 | Testing only |
| Basic | 150 | $6/mo (annual) | Hobby / light use |
| Pro | 700 | $27/mo (annual) | Regular creators |
| Ultimate | 1,500 | $55/mo (annual) | Daily production |
Commercial rights: Basic and above.
Watermarks: Free tier only.
Priority processing: Pro and above.
Plan names, prices, and credit allowances above are hand-maintained and not verifiable from the API catalog (last reviewed 2026-07-06, not re-verified against the live UI) — check higgsfield.ai/pricing before quoting them.
Credit Cost Tiers (Approximate)
Model roster reviewed against the 2026-07-05 catalog snapshot; tier placements are hand-maintained — verify live before quoting.
Low cost: Seedance 2.0 Fast / Mini, standard image generation, Nano Banana 2 Lite Medium cost: Kling 2.6 (legacy), Kling 3.0 Turbo, Wan 2.6/2.7 (and 2.5 — not in the API catalog, 2026-09-26 — verify in the live UI), Minimax Hailuo 2.3, standard I2V High cost: Kling 3.0 (pro/4K modes), Seedance 2.0 at 1080p/4K, Veo 3 / 3.1, Cinema Studio Apps: Vary widely — one-click apps are generally efficient
"Seedance Pro" is a legacy UI label — not in the API catalog (2026-07-05); its budget slot is now Seedance 2.0 Fast / Mini. Sora 2 is retired — OpenAI shut the Sora 2 API down on 2026-09-24 and Higgsfield UI availability is unconfirmed; don't recommend it (
../../model-guide.md).
Quote From the Ledger, Not From Vibes
Before quoting any credit estimate for multi-shot work, run the generation ledger and cite the numbers:
python3 ../../scripts/higgsfield_memory.py ratio <project> --credits
python3 ../../scripts/higgsfield_memory.py budget <project> --shots <manifest.json>ratiogives empirical takes-per-kept per shot type, with the structural-vs-stochastic rejection split (high structural% = rewrite the prompt, don't re-roll; high stochastic% = priced re-roll territory).budgetmultiplies a planned shot manifest by those ratios → expected generations + credit estimate with a stated confidence level.- Never budget from a row marked
low-n(under 5 logged generations) — the tool flags them; respect the flag. - If the ledger is empty or thin, say so explicitly and use the
documented default planning ratios — 2–3:1 simple shots, 4–6:1 complex
shots — labeled as defaults, not data. The
budgetcommand does this labeling automatically; keep the label when you relay the estimate. - Every logged generation sharpens these numbers — the logging workflow is
one command (
../higgsfield-recall/SKILL.md§ Log the Generation Result).
The 5 Most Common Credit Waste Patterns
1. Generating video before perfecting the image The single biggest waste. If your Hero Frame (base image) isn't right, every animated version will be wrong too. Fix: Spend extra time on image generation (low cost) → animate once (higher cost)
2. Long prompts that fight each other Over-specified prompts create conflicting instructions, forcing multiple regenerations. Fix: Under-specialize on elements you don't care about. Specify only what matters.
3. Changing multiple variables between generations If you change the prompt, the model, AND the camera in one go, you can't learn what fixed what. Fix: Change one thing at a time. Systematic iteration is faster than random retries.
4. Using premium tiers (Kling 3.0 pro/4K, Seedance 2.0 4K, Veo 3.1) for simple shots
Premium models for simple single-character, single-camera shots.
Fix: Reserve premium models for scenes that genuinely need their capabilities.
Kling 3.0 Turbo or Kling 2.6 (legacy) handles most character drama at lower cost —
and on Kling 3.0, sound: off gives a silent video at lower credits (per the
live spec). Seedance has the same switch (generate_audio: false).
5. Not using Apps for tasks Apps are built for Face swap, product placement, style transfer — doing these manually via prompt takes more credits than the App designed for that task. Fix: Check the Apps library first. If an App covers your use case, use it.
The Hero Frame Efficiency Method
This is the single highest-leverage credit optimization technique:
Step 1: Generate 5–10 image variations (very low credit cost)
→ Find the one that's closest to your vision
Step 2: Refine that one image with inpainting/editing (low cost)
→ Get it exactly right
Step 3: Animate ONCE from the perfect Hero Frame (medium-high cost)
→ First animation attempt is already working with a strong foundationResult: You spend more on cheap image credits, far less on expensive video credits. The credit math almost always favors this approach.
Model Selection by Budget Scenario
Tight budget (Basic plan — 150 credits):
- Primary model: Seedance 2.0 Fast / Mini (fast, low cost; the old "Seedance Pro" is a legacy UI label — not in the API catalog)
- Character shots: Kling 3.0 Turbo (or legacy Kling 2.6) only when quality requires it
- Avoid: Kling 3.0 pro/4K modes, Seedance 2.0 at 1080p/4K, Veo 3 / 3.1
- Strategy: Use Apps heavily — they're credit-efficient for their use cases
Mid budget (Pro plan — 700 credits):
- Primary models: Kling 3.0 Turbo, Kling 2.6 (legacy), Wan 2.7, Minimax Hailuo 2.3 (Wan 2.5: not in the API catalog, 2026-09-26 — verify in the live UI)
- Reserve Kling 3.0 (pro/4K) / Seedance 2.0 4K for hero shots only (Sora 2 is retired — OpenAI shut its API down 2026-09-24; don't budget for it)
- Use Cinema Studio for your two or three most important scenes
- Strategy: Iterate in image first, commit in video second
High volume (Ultimate — 1,500 credits):
- Full access to all models
- Cinema Studio as primary workflow for quality content
- Audio is no longer a one-model feature — Seedance 2.0 / 2.0 Mini / 1.5 Pro
generate native audio (
generate_audio, default on for 2.0), Kling 3.0 and 2.6 have asoundswitch, Veo 3.1 Lite hasgenerate_audio. Pick by scene fit, then toggle audio — don't pick the model for the audio. - Strategy: Invest in Moodboard + Soul ID upfront to avoid style drift
Platform Efficiency Tips
Use presets before writing from scratch Higgsfield's presets (visual styles, motion presets, Cinema Studio genres) encode a lot of quality that's hard to replicate with text alone. Always start with a preset as a base, then customize.
Check the Community gallery before generating Before burning credits on a new style or effect you haven't tried, find a community example that uses it. See what actually works before committing.
Use Assist for quick decisions "Should I use Kling 3.0 or Seedance 2.0 for this?" → ask Assist in 5 seconds rather than generating two test clips.
Save successful prompts When a generation works well, save the complete prompt immediately. Higgsfield doesn't have a native prompt library — you need your own. A simple text file organized by genre works well.
Chain Apps with video for social content Generate a base clip with Kling 2.6, then feed it through an App (Transitions, Style Snap, Urban Cuts) for the final social-ready version. Two steps, total cost is still lower than generating a "perfect" clip from scratch.
Batch similar shots together If you're using the same Soul ID character in 5 different scenes, generate them in the same session. The Hero Frame warm-up time is essentially zero if you're using the same Reference Anchor.
The Platform Learning Path (Credit-Efficient)
Week 1 — Image foundation (Basic plan)
- Master Soul 2.0 + Nano Banana Pro for image generation
- Build your first Soul ID character
- Create a Moodboard for your project
- Target: consistently generating images you're proud of
Week 2 — Simple video (Basic plan)
- I2V with Kling 2.6, single camera control, one scene type
- Target: one good 5-second clip you'd actually use
Week 3 — Cinema Studio (Pro plan)
- One Cinema Studio project, 3–5 shots
- Learn the Hero Frame + Reference Anchor workflow
- Target: a 15–20 second sequence with consistent character
Week 4+ — Full production (Pro or Ultimate)
- Mix Cinema Studio with Apps for efficiency
- Add Moodboard + Soul ID for consistent series content
- Use Higgsfield Assist for rapid iteration
Related skills
higgsfield-models— Detailed model comparison beyond what Assist provideshiggsfield-prompt— MCSLA formula for structured prompt buildinghiggsfield-apps— Apps Assist can recommendhiggsfield-pipeline— Full production workflows
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

