What Advanced Settings Actually Do
Every Agent in Ongage Studio comes with a set of Advanced Configuration options. Most of the time, these are pre-filled with defaults that work well for your content type. You can leave them as-is and never touch a thing.
But if you’re curious — or working with more advanced users or edge cases — it’s helpful to understand what these settings actually control.
This section explains the most common advanced options:
- Temperature
- LLM Model
- Output Format
- Image Model
- Token Limits (rare)
- Fallbacks and Input Behaviors
Again, none of this is required to generate great content. But if you ever want to fine-tune how your Agent behaves, this is where it happens.
Temperature
What it does:
Controls how “creative” or “safe” the AI model is allowed to be. Technically, it adjusts the randomness in word selection.
| Setting | Behavior | Best for |
|---|---|---|
| 0.0 – 0.3 | Very stable, repeatable, low-risk output | SEO tags, metadata, summaries |
| 0.4 – 0.7 | Balanced, natural, and generally coherent | Blog posts, product descriptions |
| 0.8 – 1.0 | More expressive and variable | Headlines, storytelling, ad copy |
Studio default: Most Agents are set between 0.6 and 0.9 depending on the content type..
What it does:
Determines which large language model (LLM) is used when the Agent runs. Each model has its own strengths in tone, reasoning, speed, formatting, or style. You don’t need to choose one manually — defaults are already set per Agent type — but this setting is visible if you want to adjust it.
Here’s how the current options compare:
General-Purpose Models
| Model | Best Uses | What Makes It Stand Out |
|---|---|---|
| GPT-4.1 | All-purpose writing | Strong reasoning, great tone control, structured |
| GPT-4.1 Mini | Short-form, light articles | Fast and cost-efficient with decent tone |
| GPT-4.1 Nano | Microtasks, utilities, simple outputs | Cheapest option, best for low-stakes tasks |
| GPT-4o | On-brand formatting, clean copy | Speedy, polished, good for summaries or polish |
| GPT-4o Mini | Lightweight content, speed runs | Near-instant, slightly less nuance |
| Claude 3.7 Sonnet | Articles, summaries, editorial content | Smooth, human-like tone with good structure |
| Claude 3.5 Sonnet | General content, clean prose | Balanced clarity and friendliness |
| Claude 3.5 Haiku | Quick summaries, short outputs | Compact, fast, and readable |
| Claude 4.0 Sonnet | High-end longform writing | Deep tone nuance and advanced formatting |
| phi-4-multimodal | Text + vision tasks | Combines images and text — experimental use |
Cost-Efficient / Bulk Writing Models
| Model | Best Uses | What Makes It Stand Out |
|---|---|---|
| O1 / O1 Mini | SEO content, listicles, repeated formats | Fast, scalable, good formatting at low cost |
| O3 Mini | Rewrites, summaries, light edits | Efficient, simple output, limited tone nuance |
Contextual & Research-Heavy Models
| Model | Best Uses | What Makes It Stand Out |
|---|---|---|
| Perplexity Sonar Small | Quick fact-checking, short answers | Lightweight, fast context awareness |
| Perplexity Sonar Large | News-based content, general accuracy | Well-grounded with good recall |
| Perplexity Sonar Huge | Deep research, factual recall, citations | Maximum reference memory — best for knowledge tasks |
Experimental & Specialized Models
| Model | Best Uses | What Makes It Stand Out |
|---|---|---|
| FLUX.1 Kontext Max | Domain-specific editorial tasks | Deep logic + style awareness (experimental) |
| FLUX.1 Kontext Pro | Precision tasks with structure and tone | Semantically sharp, ideal for structured logic |
| flux-1.1-pro | High-quality tone-aware writing | Best for tone-sensitive longform or branded work |
| flux-1.1-pro-ultra | Polished precision editorial | Cleanest output with high accuracy (gated use) |
| Bytedance Seedream-3 | Creative copywriting, quirky ideas | Stylistic flair with unconventional output |
Image Generation Models
| Model | Best Uses | What Makes It Stand Out |
|---|---|---|
| Ideogram v3 Turbo | Concept drafts, social posts | Fastest rendering, good for quick visuals |
| Ideogram v3 Quality | Final images, ad visuals | Best resolution and clarity |
| Ideogram v3 Balanced | General use, branded graphics | Balanced speed and detail |
| Google Imagen 4 | Realistic editorial images | Top-tier realism and accuracy |
| Bytedance Seedream-3 | Surreal or stylized visuals | Artistic, dreamlike outputs |
| Flux Kontext Max (Image) | Data visuals, diagrams, structure-heavy | Great for labeled or logic-heavy content |
| Flux Kontext Pro (Image) | Detail-rich illustrations | Consistent, high-res images with structure |
| flux-1.1-pro (Image) | UI images, charts, diagrams | Clean, professional formatting |
| phi-4-multimodal-instruct | Image + instruction tasks | Combined visual and textual understanding |
How to Choose (If You Ever Need To)
Most users will never change the model setting. But if you're experimenting, here's a simple guide:
- Use GPT-4.1 or Claude 3.7 for general-purpose writing and articles
- Use O1/O3 for high-volume, low-cost content (e.g. SEO, product listings)
- Use Perplexity for fact-based or reference-rich generation
- Use FLUX or Seedream only if you’re testing tone-specific or edge-case prompts
Default Behavior
Each Agent is pre-configured with the model best suited to its task. These defaults are carefully selected based on:
- Output type (article, headline, SEO block, etc.)
- Tone and format complexity
- Brand Voice style (where applicable)
- Balance of speed, quality, and cost
If you open the Advanced Configuration panel, you’ll see the selected model. You can change it — but unless you know exactly what you want to test, you don’t need to.
Output Format
What it does:
Defines how the output is returned and formatted. This matters when integrating content into websites, feeds, or export pipelines.
Options:
- HTML – Default for articles, headlines, SEO, CTAs
- XML – Used for structured publishing, e.g. RSS feeds
- JSON – Used for programmatic integration with other tools
- Plain Text – Used in test environments or raw outputs
Default behavior: Most Agents are set to HTML. You can toggle this only if your workflow requires a specific format.
Image Model (only for image-generating Agents)
What it controls:
Which AI model is used to generate images. This is only visible when you use Agents that include visual output.
| Model | Best For |
|---|---|
| Ideogram | Stylized or branded illustrations |
| Imagen 4 (Google) | Clean, photorealistic images |
| Flux | Balanced editorial visual output |
| Image Seedream 3 | Artistic, emotionally rich images with subtle surrealism |
Studio default: Auto-selected per Agent based on image needs.
Token Limit (Advanced)
What it controls:
Maximum length of input/output, measured in tokens (not words).
100 tokens ≈ 75 words.
Usually only relevant for custom-integrated or high-volume use cases.
Default behavior: Pre-set by Studio. No need to change this unless instructed.
Fallback Behavior (When Inputs Are Blank)
Studio Agents are designed to run smoothly, even if some optional inputs aren’t filled in.
| Placeholder | Fallback Behavior |
|---|---|
{{BrandVoice}} | Uses Site default Brand Voice |
{{Input_Source}} | Pulls directly from raw text, article, or topic |
{{CTAGoal}} | Defaults to “Learn More” |
Summary
You don’t need to change advanced settings — but if you want to, here’s what they control:
- Temperature adjusts creativity
- Model affects tone, structure, and cost
- Output Format sets structure (HTML, XML, etc.)
- Image Model applies only to visual content
- Token Limits rarely apply unless you’re a developer
- Fallbacks make sure your Agent still runs when inputs are missing
If you're building high-volume, client-facing, or highly tuned content systems, these are worth understanding. Otherwise, Studio will take care of them for you.